RealX blind region diagnosis + Grounding + SAM v1

推理输入只有 opaque target 与 4 个 pose-nearest context。VLM 输出语义短语，Grounding DINO 负责 box，SAM 2.1 负责 mask。Reference 在 35 条 VLM JSONL 锁定后才由 evaluator 打开。

协议与限制

这是 descriptive pilot。paired-reference mask 是 pseudo-GT，不是人工语义 GT；它可能包含轻微配准、阴影、反射和曝光差。开发 smoke case 曾用于固定 prompt，未继续按 evaluation 结果调阈值。原始 prompt、schema、subagent 审计 prompt、完整协议和所有结构化输出均随报告保存。

VLM prompt · schema · subagent prompt · protocol · summary JSON · per-frame JSON

结果摘要

33.3%positive frame recall*
20.0%clean-test frame FPR*
0.095mean raw-mask IoU
0.000median raw-mask IoU
0.101mean SafetyRecall
0.899mean Leakage
0.103mean routing-mask IoU
0.138mean box GT coverage
0.189mean box area precision
0VLM parse failures

* frame detected = raw predicted mask area > 0.5% of image.

端到端漏斗（30 positive frames）

VLM regions 73 → eligible 30（rejected 43）→ frames with eligible region 15 → frames with box 10 → detected masks 10 → IoU≥0.25 4。

0001 b_d5be48474254

train · IoU 0.567 · recall 0.665
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_1aede9f60837
b_a38e3ce498d5
b_63ebbf87a7c1
b_442bffc55440
Dice 0.723Precision 0.793SafetyRecall 0.665BF@5 0.124Box coverage 0.798Overmask 0.010
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_1aede9f60837", "observation": "red umbrella partially obscured"},
      {"context_id": "b_a38e3ce498d5", "observation": "red umbrella fully visible"},
      {"context_id": "b_63ebbf87a7c1", "observation": "red umbrella partially obscured"},
      {"context_id": "b_442bffc5540", "observation": "red umbrella fully visible"}
    ],
    "uncertainty": "red umbrella's visibility varies across views",
    "fallback_box_norm_xyxy": [292, 275, 400, 370]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "black and white umbrella",
    "grounding_queries": ["black and white umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_1aede9f60837", "observation": "black and white umbrella partially obscured"},
      {"context_id": "b_a38e3ce498d5", "observation": "black and white umbrella fully visible"},
      {"context_id": "b_63ebbf87a7c1", "observation": "black and white umbrella partially obscured"},
      {"context_id": "b_442bffc5540", "observation": "black and white umbrella fully visible"}
    ],
    "uncertainty": "black and white umbrella's visibility varies across views",
    "fallback_box_norm_xyxy": [1, 320, 170, 448]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "mannequin shirt",
    "grounding_queries": ["mannequin shirt"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_1aede9f60837", "observation": "mannequin shirt partially obscured"},
      {"context_id": "b_a38e3ce498d5", "observation": "mannequin shirt fully visible"},
      {"context_id": "b_63ebbf87a7c1", "observation": "mannequin shirt partially obscured"},
      {"context_id": "b_442bffc5540", "observation": "mannequin shirt fully visible"}
    ],
    "uncertainty": "mannequin shirt's visibility varies across views",
    "fallback_box_norm_xyxy": [170, 280, 290, 448]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_d5be48474254",
    "target_sha256": "c8e0464012d765884090d74c36c513738c8d3ff0de56534b10d97666e0dfae81",
    "context_ids": [
      "b_1aede9f60837",
      "b_a38e3ce498d5",
      "b_63ebbf87a7c1",
      "b_442bffc55440"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_1aede9f60837",
            "observation": "red umbrella partially obscured"
          },
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "red umbrella fully visible"
          },
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "red umbrella partially obscured"
          },
          {
            "context_id": "b_442bffc5540",
            "observation": "red umbrella fully visible"
          }
        ],
        "uncertainty": "red umbrella's visibility varies across views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "grounding_queries": [
          "black and white umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_1aede9f60837",
            "observation": "black and white umbrella partially obscured"
          },
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "black and white umbrella fully visible"
          },
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "black and white umbrella partially obscured"
          },
          {
            "context_id": "b_442bffc5540",
            "observation": "black and white umbrella fully visible"
          }
        ],
        "uncertainty": "black and white umbrella's visibility varies across views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "mannequin shirt",
        "grounding_queries": [
          "mannequin shirt"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_1aede9f60837",
            "observation": "mannequin shirt partially obscured"
          },
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "mannequin shirt fully visible"
          },
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "mannequin shirt partially obscured"
          },
          {
            "context_id": "b_442bffc5540",
            "observation": "mannequin shirt fully visible"
          }
        ],
        "uncertainty": "mannequin shirt's visibility varies across views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_d5be48474254",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "queries": [
          "red umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            113.717529296875,
            702.1360473632812,
            436.87554931640625,
            1007.55078125
          ],
          [
            753.3574829101562,
            598.4653930664062,
            1055.838623046875,
            894.8519287109375
          ]
        ],
        "box_scores": [
          0.3146287798881531,
          0.2932315170764923
        ],
        "box_queries": [
          "red umbrella",
          "red umbrella"
        ],
        "sam_predicted_iou": [
          0.99609375,
          0.97265625
        ],
        "mask_path": "masks/b_d5be48474254_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "queries": [
          "black and white umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            113.71705627441406,
            701.9348754882812,
            437.05767822265625,
            1007.9161376953125
          ]
        ],
        "box_scores": [
          0.27239349484443665
        ],
        "box_queries": [
          "black and white umbrella"
        ],
        "sam_predicted_iou": [
          0.99609375
        ],
        "mask_path": "masks/b_d5be48474254_r2.png"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "mannequin shirt",
        "queries": [
          "mannequin shirt"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_d5be48474254_r3.png"
      }
    ],
    "final_mask_path": "masks/b_d5be48474254.png",
    "routing_mask_path": "masks/b_d5be48474254_routing.png",
    "predicted_fraction": 0.045033856966976794
  }
}

0002 b_1aede9f60837

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_d5be48474254
b_a38e3ce498d5
b_63ebbf87a7c1
b_442bffc55440
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_d5be48474254", "observation": "red umbrella is visible"},
      {"context_id": "b_a38e3ce498d5", "observation": "red umbrella is visible"},
      {"context_id": "b_63ebbf87a7c1", "observation": "red umbrella is visible"},
      {"context_id": "b_442bffc55440", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [280, 258, 380, 348]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "black and white umbrella",
    "grounding_queries": ["black and white umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_d5be48474254", "observation": "black and white umbrella is visible"},
      {"context_id": "b_a38e3ce498d5", "observation": "black and white umbrella is visible"},
      {"context_id": "b_63ebbf87a7c1", "observation": "black and white umbrella is visible"},
      {"context_id": "b_442bffc55440", "observation": "black and white umbrella is visible"}
    ],
    "uncertainty": "black and white umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0, 138, 280, 476]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "mannequin shirt",
    "grounding_queries": ["mannequin shirt"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_a38e3ce498d5", "observation": "mannequin shirt is visible"},
      {"context_id": "b_63ebbf87a7c1", "observation": "mannequin shirt is visible"},
      {"context_id": "b_442bffc55440", "observation": "mannequin shirt is visible"}
    ],
    "uncertainty": "mannequin shirt is consistently visible across two context images",
    "fallback_box_norm_xyxy": [168, 308, 310, 476]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_1aede9f60837",
    "target_sha256": "8fa7c6b67704a3493ad59f7ce40137e6dd08cf91d93521af022d99e65c613084",
    "context_ids": [
      "b_d5be48474254",
      "b_a38e3ce498d5",
      "b_63ebbf87a7c1",
      "b_442bffc55440"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_d5be48474254",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "grounding_queries": [
          "black and white umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_d5be48474254",
            "observation": "black and white umbrella is visible"
          },
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "black and white umbrella is visible"
          },
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "black and white umbrella is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "black and white umbrella is visible"
          }
        ],
        "uncertainty": "black and white umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "mannequin shirt",
        "grounding_queries": [
          "mannequin shirt"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "mannequin shirt is visible"
          },
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "mannequin shirt is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "mannequin shirt is visible"
          }
        ],
        "uncertainty": "mannequin shirt is consistently visible across two context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_1aede9f60837",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_1aede9f60837.png",
    "routing_mask_path": "masks/b_1aede9f60837_routing.png",
    "predicted_fraction": 0.0
  }
}

0003 b_a38e3ce498d5

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_63ebbf87a7c1
b_1aede9f60837
b_442bffc55440
b_6c74696a698a
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "mannequin with shirt",
    "grounding_queries": ["mannequin with shirt"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_63ebbf87a7c1", "observation": "mannequin with shirt is visible"},
      {"context_id": "b_1aede9f60837", "observation": "mannequin with shirt is visible"},
      {"context_id": "b_442bffc55440", "observation": "mannequin with shirt is visible"},
      {"context_id": "b_6c74696a698a", "observation": "mannequin with shirt is visible"}
    ],
    "uncertainty": "mannequin with shirt is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_63ebbf87a7c1", "observation": "red umbrella is visible"},
      {"context_id": "b_1aede9f60837", "observation": "red umbrella is visible"},
      {"context_id": "b_442bffc55440", "observation": "red umbrella is visible"},
      {"context_id": "b_6c74696a698a", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "sign with Chinese characters",
    "grounding_queries": ["sign with Chinese characters"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_63ebbf87a7c1", "observation": "sign with Chinese characters is visible"},
      {"context_id": "b_1aede9f60837", "observation": "sign with Chinese characters is visible"},
      {"context_id": "b_442bffc55440", "observation": "sign with Chinese characters is visible"},
      {"context_id": "b_6c74696a698a", "observation": "sign with Chinese characters is visible"}
    ],
    "uncertainty": "sign with Chinese characters is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_a38e3ce498d5",
    "target_sha256": "1f9ba7f01a8ec6049827973e61b97173ff02690eff8dcb5da68e289ff5445ea2",
    "context_ids": [
      "b_63ebbf87a7c1",
      "b_1aede9f60837",
      "b_442bffc55440",
      "b_6c74696a698a"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mannequin with shirt",
        "grounding_queries": [
          "mannequin with shirt"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "mannequin with shirt is visible"
          },
          {
            "context_id": "b_1aede9f60837",
            "observation": "mannequin with shirt is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "mannequin with shirt is visible"
          },
          {
            "context_id": "b_6c74696a698a",
            "observation": "mannequin with shirt is visible"
          }
        ],
        "uncertainty": "mannequin with shirt is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_1aede9f60837",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_6c74696a698a",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "sign with Chinese characters",
        "grounding_queries": [
          "sign with Chinese characters"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "sign with Chinese characters is visible"
          },
          {
            "context_id": "b_1aede9f60837",
            "observation": "sign with Chinese characters is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "sign with Chinese characters is visible"
          },
          {
            "context_id": "b_6c74696a698a",
            "observation": "sign with Chinese characters is visible"
          }
        ],
        "uncertainty": "sign with Chinese characters is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_a38e3ce498d5",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_a38e3ce498d5.png",
    "routing_mask_path": "masks/b_a38e3ce498d5_routing.png",
    "predicted_fraction": 0.0
  }
}

0004 b_63ebbf87a7c1

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_442bffc55440
b_a38e3ce498d5
b_6c74696a698a
b_e21902361190
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_442bffc55440", "observation": "umbrella is visible"},
      {"context_id": "b_a38e3ce498d5", "observation": "umbrella is visible"},
      {"context_id": "b_6c74696a698a", "observation": "umbrella is visible"},
      {"context_id": "b_e21902361190", "observation": "umbrella is visible"}
    ],
    "uncertainty": "umbrella is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "sign with Chinese characters",
    "grounding_queries": ["sign with Chinese characters"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_442bffc55440", "observation": "sign is visible"},
      {"context_id": "b_a38e3ce498d5", "observation": "sign is visible"},
      {"context_id": "b_6c74696a698a", "observation": "sign is visible"},
      {"context_id": "b_e21902361190", "observation": "sign is visible"}
    ],
    "uncertainty": "sign is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_63ebbf87a7c1",
    "target_sha256": "2e7c8d2413cc7d1ba3ba9cc698dc06124e65069ca702187a418ca895b8afc20a",
    "context_ids": [
      "b_442bffc55440",
      "b_a38e3ce498d5",
      "b_6c74696a698a",
      "b_e21902361190"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_442bffc55440",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_6c74696a698a",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_e21902361190",
            "observation": "umbrella is visible"
          }
        ],
        "uncertainty": "umbrella is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "sign with Chinese characters",
        "grounding_queries": [
          "sign with Chinese characters"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_442bffc55440",
            "observation": "sign is visible"
          },
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "sign is visible"
          },
          {
            "context_id": "b_6c74696a698a",
            "observation": "sign is visible"
          },
          {
            "context_id": "b_e21902361190",
            "observation": "sign is visible"
          }
        ],
        "uncertainty": "sign is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_63ebbf87a7c1",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_63ebbf87a7c1.png",
    "routing_mask_path": "masks/b_63ebbf87a7c1_routing.png",
    "predicted_fraction": 0.0
  }
}

0005 b_6c74696a698a

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_c7168d6511cd
b_63ebbf87a7c1
b_e21902361190
b_442bffc55440
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_c7168d6511cd", "observation": "red umbrella is visible"},
      {"context_id": "b_63ebbf87a7c1", "observation": "red umbrella is visible"},
      {"context_id": "b_e21902361190", "observation": "red umbrella is visible"},
      {"context_id": "b_442bffc55440", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [278, 250, 382, 340]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "white sign with red text",
    "grounding_queries": ["white sign with red text"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_c7168d6511cd", "observation": "white sign with red text is visible"},
      {"context_id": "b_63ebbf87a7c1", "observation": "white sign with red text is visible"},
      {"context_id": "b_e21902361190", "observation": "white sign with red text is visible"},
      {"context_id": "b_442bffc55440", "observation": "white sign with red text is visible"}
    ],
    "uncertainty": "white sign with red text is consistently visible across all context images",
    "fallback_box_norm_xyxy": [170, 250, 234, 300]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_6c74696a698a",
    "target_sha256": "692c1ff802d24d43d30ea964cbb5401b228209287c7e3cb45fe06d378039eadc",
    "context_ids": [
      "b_c7168d6511cd",
      "b_63ebbf87a7c1",
      "b_e21902361190",
      "b_442bffc55440"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_c7168d6511cd",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_e21902361190",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "white sign with red text",
        "grounding_queries": [
          "white sign with red text"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_c7168d6511cd",
            "observation": "white sign with red text is visible"
          },
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "white sign with red text is visible"
          },
          {
            "context_id": "b_e21902361190",
            "observation": "white sign with red text is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "white sign with red text is visible"
          }
        ],
        "uncertainty": "white sign with red text is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_6c74696a698a",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_6c74696a698a.png",
    "routing_mask_path": "masks/b_6c74696a698a_routing.png",
    "predicted_fraction": 0.0
  }
}

0006 b_c7168d6511cd

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_6c74696a698a
b_e21902361190
b_f2d05c389498
b_f9646cc9c954
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_6c74696a698a", "observation": "red umbrella is visible"},
      {"context_id": "b_e21902361190", "observation": "red umbrella is visible"},
      {"context_id": "b_f2d05c389498", "observation": "red umbrella is visible"},
      {"context_id": "b_f9646cc9c954", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "black and white umbrella",
    "grounding_queries": ["black and white umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_f9646cc9c954", "observation": "black and white umbrella is visible"}
    ],
    "uncertainty": "black and white umbrella is only visible in one context image",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_c7168d6511cd",
    "target_sha256": "a6ed36e8527621f968cb684b2277d0e217f8744eec4d5bbc6be1c9a93d1e5b11",
    "context_ids": [
      "b_6c74696a698a",
      "b_e21902361190",
      "b_f2d05c389498",
      "b_f9646cc9c954"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_6c74696a698a",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_e21902361190",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_f2d05c389498",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_f9646cc9c954",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "grounding_queries": [
          "black and white umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_f9646cc9c954",
            "observation": "black and white umbrella is visible"
          }
        ],
        "uncertainty": "black and white umbrella is only visible in one context image",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_c7168d6511cd",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "queries": [
          "black and white umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_c7168d6511cd_r2.png"
      }
    ],
    "final_mask_path": "masks/b_c7168d6511cd.png",
    "routing_mask_path": "masks/b_c7168d6511cd_routing.png",
    "predicted_fraction": 0.0
  }
}

0007 b_f9646cc9c954

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_f2d05c389498
b_1ea940c8ac4e
b_39a6a990391f
b_c7168d6511cd
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "black and white umbrella",
    "grounding_queries": ["black and white umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_f2d05c389498", "observation": "umbrella is visible"},
      {"context_id": "b_1ea940c8ac4e", "observation": "umbrella is visible"},
      {"context_id": "b_39a6a990391f", "observation": "umbrella is visible"},
      {"context_id": "b_c7168d6511cd", "observation": "umbrella is visible"}
    ],
    "uncertainty": "umbrella is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_f2d05c389498", "observation": "umbrella is visible"},
      {"context_id": "b_1ea940c8ac4e", "observation": "umbrella is visible"},
      {"context_id": "b_39a6a990391f", "observation": "umbrella is visible"},
      {"context_id": "b_c7168d6511cd", "observation": "umbrella is visible"}
    ],
    "uncertainty": "umbrella is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "flower arrangement",
    "grounding_queries": ["flower arrangement"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_f2d05c389498", "observation": "flower arrangement is visible"},
      {"context_id": "b_1ea940c8ac4e", "observation": "flower arrangement is visible"},
      {"context_id": "b_39a6a990391f", "observation": "flower arrangement is visible"},
      {"context_id": "b_c7168d6511cd", "observation": "flower arrangement is visible"}
    ],
    "uncertainty": "flower arrangement is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_f9646cc9c954",
    "target_sha256": "0f951d7fb7e2f9b4f41f6adca3e1ca2cb66f91a97ab2476d57248824398134e6",
    "context_ids": [
      "b_f2d05c389498",
      "b_1ea940c8ac4e",
      "b_39a6a990391f",
      "b_c7168d6511cd"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "grounding_queries": [
          "black and white umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_f2d05c389498",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_39a6a990391f",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_c7168d6511cd",
            "observation": "umbrella is visible"
          }
        ],
        "uncertainty": "umbrella is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_f2d05c389498",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_39a6a990391f",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_c7168d6511cd",
            "observation": "umbrella is visible"
          }
        ],
        "uncertainty": "umbrella is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "flower arrangement",
        "grounding_queries": [
          "flower arrangement"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_f2d05c389498",
            "observation": "flower arrangement is visible"
          },
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "flower arrangement is visible"
          },
          {
            "context_id": "b_39a6a990391f",
            "observation": "flower arrangement is visible"
          },
          {
            "context_id": "b_c7168d6511cd",
            "observation": "flower arrangement is visible"
          }
        ],
        "uncertainty": "flower arrangement is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_f9646cc9c954",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_f9646cc9c954.png",
    "routing_mask_path": "masks/b_f9646cc9c954_routing.png",
    "predicted_fraction": 0.0
  }
}

0008 b_1ea940c8ac4e

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_39a6a990391f
b_31faa8f0af47
b_f9646cc9c954
b_0ed8205374dd
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_39a6a990391f", "observation": "red umbrella is visible"},
      {"context_id": "b_31faa8f0af47", "observation": "red umbrella is visible"},
      {"context_id": "b_f9646cc9c954", "observation": "red umbrella is visible"},
      {"context_id": "b_0ed8205374dd", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "black umbrella",
    "grounding_queries": ["black umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_39a6a990391f", "observation": "black umbrella is visible"},
      {"context_id": "b_31faa8f0af47", "observation": "black umbrella is visible"},
      {"context_id": "b_f9646cc9c954", "observation": "black umbrella is visible"},
      {"context_id": "b_0ed8205374dd", "observation": "black umbrella is visible"}
    ],
    "uncertainty": "black umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "green wall",
    "grounding_queries": ["green wall"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_39a6a990391f", "observation": "green wall is visible"},
      {"context_id": "b_31faa8f0af47", "observation": "green wall is visible"},
      {"context_id": "b_f9646cc9c954", "observation": "green wall is visible"},
      {"context_id": "b_0ed8205374dd", "observation": "green wall is visible"}
    ],
    "uncertainty": "green wall is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_1ea940c8ac4e",
    "target_sha256": "5939628718d60686b21ce724529ae903572b899fa39971c25aa12f559a30b34d",
    "context_ids": [
      "b_39a6a990391f",
      "b_31faa8f0af47",
      "b_f9646cc9c954",
      "b_0ed8205374dd"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_39a6a990391f",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_31faa8f0af47",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_f9646cc9c954",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_0ed8205374dd",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "black umbrella",
        "grounding_queries": [
          "black umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_39a6a990391f",
            "observation": "black umbrella is visible"
          },
          {
            "context_id": "b_31faa8f0af47",
            "observation": "black umbrella is visible"
          },
          {
            "context_id": "b_f9646cc9c954",
            "observation": "black umbrella is visible"
          },
          {
            "context_id": "b_0ed8205374dd",
            "observation": "black umbrella is visible"
          }
        ],
        "uncertainty": "black umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "green wall",
        "grounding_queries": [
          "green wall"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_39a6a990391f",
            "observation": "green wall is visible"
          },
          {
            "context_id": "b_31faa8f0af47",
            "observation": "green wall is visible"
          },
          {
            "context_id": "b_f9646cc9c954",
            "observation": "green wall is visible"
          },
          {
            "context_id": "b_0ed8205374dd",
            "observation": "green wall is visible"
          }
        ],
        "uncertainty": "green wall is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_1ea940c8ac4e",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_1ea940c8ac4e.png",
    "routing_mask_path": "masks/b_1ea940c8ac4e_routing.png",
    "predicted_fraction": 0.0
  }
}

0009 b_31faa8f0af47

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_0ed8205374dd
b_1ea940c8ac4e
b_39a6a990391f
b_57cf54a3bbc0
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "black curtain",
    "grounding_queries": ["black curtain"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_0ed8205374dd", "observation": "black curtain is absent"},
      {"context_id": "b_1ea940c8ac4e", "observation": "black curtain is absent"},
      {"context_id": "b_39a6a990391f", "observation": "black curtain is absent"},
      {"context_id": "b_57cf54a3bbc0", "observation": "black curtain is absent"}
    ],
    "uncertainty": "black curtain is consistently absent across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_31faa8f0af47",
    "target_sha256": "8522b8e9cf92d396f1857e8197dbba361ef53ce127af536586946128b5e576d0",
    "context_ids": [
      "b_0ed8205374dd",
      "b_1ea940c8ac4e",
      "b_39a6a990391f",
      "b_57cf54a3bbc0"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black curtain",
        "grounding_queries": [
          "black curtain"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_0ed8205374dd",
            "observation": "black curtain is absent"
          },
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "black curtain is absent"
          },
          {
            "context_id": "b_39a6a990391f",
            "observation": "black curtain is absent"
          },
          {
            "context_id": "b_57cf54a3bbc0",
            "observation": "black curtain is absent"
          }
        ],
        "uncertainty": "black curtain is consistently absent across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_31faa8f0af47",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black curtain",
        "queries": [
          "black curtain"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_31faa8f0af47_r1.png"
      }
    ],
    "final_mask_path": "masks/b_31faa8f0af47.png",
    "routing_mask_path": "masks/b_31faa8f0af47_routing.png",
    "predicted_fraction": 0.0
  }
}

0010 b_0ed8205374dd

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_31faa8f0af47
b_57cf54a3bbc0
b_c7dd5015d023
b_1ea940c8ac4e
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "black curtain",
    "grounding_queries": ["black curtain"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_31faa8f0af47", "observation": "black curtain partially obscures the teddy bear"},
      {"context_id": "b_57cf54a3bbc0", "observation": "black curtain fully covers the teddy bear"},
      {"context_id": "b_c7dd5015d023", "observation": "black curtain is not visible"},
      {"context_id": "b_1ea940c8ac4e", "observation": "black curtain is not visible"}
    ],
    "uncertainty": "black curtain is only visible in one context image",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "brown cushion",
    "grounding_queries": ["brown cushion"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_31faa8f0af47", "observation": "brown cushion is visible"},
      {"context_id": "b_57cf54a3bbc0", "observation": "brown cushion is visible"},
      {"context_id": "b_c7dd5015d023", "observation": "brown cushion is visible"},
      {"context_id": "b_1ea940c8ac4e", "observation": "brown cushion is visible"}
    ],
    "uncertainty": "brown cushion is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_31faa8f0af47", "observation": "red umbrella is visible"},
      {"context_id": "b_57cf54a3bbc0", "observation": "red umbrella is visible"},
      {"context_id": "b_c7dd5015d023", "observation": "red umbrella is visible"},
      {"context_id": "b_1ea940c8ac4e", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_0ed8205374dd",
    "target_sha256": "9fc4d940f95eac42ee267d66fd919fd55d45c2ccc498dcf4dfece29c22bbdabd",
    "context_ids": [
      "b_31faa8f0af47",
      "b_57cf54a3bbc0",
      "b_c7dd5015d023",
      "b_1ea940c8ac4e"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black curtain",
        "grounding_queries": [
          "black curtain"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_31faa8f0af47",
            "observation": "black curtain partially obscures the teddy bear"
          },
          {
            "context_id": "b_57cf54a3bbc0",
            "observation": "black curtain fully covers the teddy bear"
          },
          {
            "context_id": "b_c7dd5015d023",
            "observation": "black curtain is not visible"
          },
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "black curtain is not visible"
          }
        ],
        "uncertainty": "black curtain is only visible in one context image",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "brown cushion",
        "grounding_queries": [
          "brown cushion"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_31faa8f0af47",
            "observation": "brown cushion is visible"
          },
          {
            "context_id": "b_57cf54a3bbc0",
            "observation": "brown cushion is visible"
          },
          {
            "context_id": "b_c7dd5015d023",
            "observation": "brown cushion is visible"
          },
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "brown cushion is visible"
          }
        ],
        "uncertainty": "brown cushion is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_31faa8f0af47",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_57cf54a3bbc0",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_c7dd5015d023",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_0ed8205374dd",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black curtain",
        "queries": [
          "black curtain"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_0ed8205374dd_r1.png"
      }
    ],
    "final_mask_path": "masks/b_0ed8205374dd.png",
    "routing_mask_path": "masks/b_0ed8205374dd_routing.png",
    "predicted_fraction": 0.0
  }
}

0011 b_c7dd5015d023

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_3fad5029fb0d
b_57cf54a3bbc0
b_0a52973d1b8d
b_0ed8205374dd
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_3fad5029fb0d", "observation": "red umbrella partially obscured"},
      {"context_id": "b_57cf54a3bbc0", "observation": "red umbrella fully visible"},
      {"context_id": "b_0a52973d1b8d", "observation": "red umbrella fully visible"},
      {"context_id": "b_0ed8205374dd", "observation": "red umbrella fully visible"}
    ],
    "uncertainty": "red umbrella consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "green curtain",
    "grounding_queries": ["green curtain"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_3fad5029fb0d", "observation": "green curtain partially obscures the teddy bear"},
      {"context_id": "b_57cf54a3bbc0", "observation": "green curtain fully visible"},
      {"context_id": "b_0a52973d1b8d", "observation": "green curtain fully visible"},
      {"context_id": "b_0ed8205374dd", "observation": "green curtain fully visible"}
    ],
    "uncertainty": "green curtain consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_c7dd5015d023",
    "target_sha256": "1a7c6617a66bee68a33a1c94e36afbcc21f1f40af2a2d54f760cdb68d0fe29c3",
    "context_ids": [
      "b_3fad5029fb0d",
      "b_57cf54a3bbc0",
      "b_0a52973d1b8d",
      "b_0ed8205374dd"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_3fad5029fb0d",
            "observation": "red umbrella partially obscured"
          },
          {
            "context_id": "b_57cf54a3bbc0",
            "observation": "red umbrella fully visible"
          },
          {
            "context_id": "b_0a52973d1b8d",
            "observation": "red umbrella fully visible"
          },
          {
            "context_id": "b_0ed8205374dd",
            "observation": "red umbrella fully visible"
          }
        ],
        "uncertainty": "red umbrella consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "green curtain",
        "grounding_queries": [
          "green curtain"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_3fad5029fb0d",
            "observation": "green curtain partially obscures the teddy bear"
          },
          {
            "context_id": "b_57cf54a3bbc0",
            "observation": "green curtain fully visible"
          },
          {
            "context_id": "b_0a52973d1b8d",
            "observation": "green curtain fully visible"
          },
          {
            "context_id": "b_0ed8205374dd",
            "observation": "green curtain fully visible"
          }
        ],
        "uncertainty": "green curtain consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_c7dd5015d023",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_c7dd5015d023.png",
    "routing_mask_path": "masks/b_c7dd5015d023_routing.png",
    "predicted_fraction": 0.0
  }
}

0012 b_3fad5029fb0d

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_6125c5000375
b_48b647235f33
b_c7dd5015d023
b_0a52973d1b8d
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "curtain",
    "grounding_queries": ["curtain"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_6125c5000375", "observation": "curtain partially visible"},
      {"context_id": "b_48b647235f33", "observation": "curtain fully visible"},
      {"context_id": "b_c7dd5015d023", "observation": "curtain fully visible"},
      {"context_id": "b_0a52973d1b8d", "observation": "curtain fully visible"}
    ],
    "uncertainty": "curtain is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_3fad5029fb0d",
    "target_sha256": "e986422f654342f73f45ef0a77076801f4c6ba007336b65b72a5b0254189af77",
    "context_ids": [
      "b_6125c5000375",
      "b_48b647235f33",
      "b_c7dd5015d023",
      "b_0a52973d1b8d"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "curtain",
        "grounding_queries": [
          "curtain"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_6125c5000375",
            "observation": "curtain partially visible"
          },
          {
            "context_id": "b_48b647235f33",
            "observation": "curtain fully visible"
          },
          {
            "context_id": "b_c7dd5015d023",
            "observation": "curtain fully visible"
          },
          {
            "context_id": "b_0a52973d1b8d",
            "observation": "curtain fully visible"
          }
        ],
        "uncertainty": "curtain is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_3fad5029fb0d",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_3fad5029fb0d.png",
    "routing_mask_path": "masks/b_3fad5029fb0d_routing.png",
    "predicted_fraction": 0.0
  }
}

0013 b_6125c5000375

train · IoU 0.872 · recall 0.872
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_3fad5029fb0d
b_48b647235f33
b_2962b6747c5c
b_0a52973d1b8d
Dice 0.932Precision 1.000SafetyRecall 0.872BF@5 0.026Box coverage 0.912Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red cloth",
    "grounding_queries": ["red cloth"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_3fad5029fb0d", "observation": "red cloth partially visible"},
      {"context_id": "b_48b647235f33", "observation": "red cloth completely absent"},
      {"context_id": "b_2962b6747c5c", "observation": "red cloth partially visible"},
      {"context_id": "b_0a52973d1b8d", "observation": "red cloth partially visible"}
    ],
    "uncertainty": "red cloth may be partially obscured in some views",
    "fallback_box_norm_xyxy": [0.0, 0.0, 0.3, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_6125c5000375",
    "target_sha256": "ceef144931ee2abde82fa3b7d28510e0c313a7be5c40c58dde0e35f1ce5d52ef",
    "context_ids": [
      "b_3fad5029fb0d",
      "b_48b647235f33",
      "b_2962b6747c5c",
      "b_0a52973d1b8d"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red cloth",
        "grounding_queries": [
          "red cloth"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_3fad5029fb0d",
            "observation": "red cloth partially visible"
          },
          {
            "context_id": "b_48b647235f33",
            "observation": "red cloth completely absent"
          },
          {
            "context_id": "b_2962b6747c5c",
            "observation": "red cloth partially visible"
          },
          {
            "context_id": "b_0a52973d1b8d",
            "observation": "red cloth partially visible"
          }
        ],
        "uncertainty": "red cloth may be partially obscured in some views",
        "fallback_box_norm_xyxy": [
          0.0,
          0.0,
          0.3,
          1.0
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_6125c5000375",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red cloth",
        "queries": [
          "red cloth"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            106.89315795898438,
            247.7123260498047,
            676.0635986328125,
            1165.3028564453125
          ]
        ],
        "box_scores": [
          0.3131753206253052
        ],
        "box_queries": [
          "red cloth"
        ],
        "sam_predicted_iou": [
          0.99609375
        ],
        "mask_path": "masks/b_6125c5000375_r1.png"
      }
    ],
    "final_mask_path": "masks/b_6125c5000375.png",
    "routing_mask_path": "masks/b_6125c5000375_routing.png",
    "predicted_fraction": 0.23016406408278398
  }
}

0014 b_2962b6747c5c

train · IoU 0.398 · recall 0.399
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_2d3a471eb3b2
b_6125c5000375
b_48b647235f33
b_42d9140f0598
Dice 0.570Precision 0.995SafetyRecall 0.399BF@5 0.068Box coverage 0.469Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "black and white umbrella",
    "grounding_queries": ["black and white umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_2d3a471eb3b2", "observation": "absent"},
      {"context_id": "b_6125c5000375", "observation": "absent"},
      {"context_id": "b_48b647235f33", "observation": "absent"},
      {"context_id": "b_42d9140f0598", "observation": "absent"}
    ],
    "uncertainty": "no other views confirm absence",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_2d3a471eb3b2", "observation": "present"},
      {"context_id": "b_6125c5000375", "observation": "absent"},
      {"context_id": "b_48b647235f33", "observation": "present"},
      {"context_id": "b_42d9140f0598", "observation": "absent"}
    ],
    "uncertainty": "no other views confirm absence",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "blue umbrella",
    "grounding_queries": ["blue umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_2d3a471eb3b2", "observation": "absent"},
      {"context_id": "b_6125c5000375", "observation": "absent"},
      {"context_id": "b_48b647235f33", "observation": "absent"},
      {"context_id": "b_42d9140f0598", "observation": "absent"}
    ],
    "uncertainty": "no other views confirm absence",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_2962b6747c5c",
    "target_sha256": "694505997ce3fd02c783fc8b4875ecd868966c155a15378d4caa53c76d6f7fef",
    "context_ids": [
      "b_2d3a471eb3b2",
      "b_6125c5000375",
      "b_48b647235f33",
      "b_42d9140f0598"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "grounding_queries": [
          "black and white umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_2d3a471eb3b2",
            "observation": "absent"
          },
          {
            "context_id": "b_6125c5000375",
            "observation": "absent"
          },
          {
            "context_id": "b_48b647235f33",
            "observation": "absent"
          },
          {
            "context_id": "b_42d9140f0598",
            "observation": "absent"
          }
        ],
        "uncertainty": "no other views confirm absence",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_2d3a471eb3b2",
            "observation": "present"
          },
          {
            "context_id": "b_6125c5000375",
            "observation": "absent"
          },
          {
            "context_id": "b_48b647235f33",
            "observation": "present"
          },
          {
            "context_id": "b_42d9140f0598",
            "observation": "absent"
          }
        ],
        "uncertainty": "no other views confirm absence",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "blue umbrella",
        "grounding_queries": [
          "blue umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_2d3a471eb3b2",
            "observation": "absent"
          },
          {
            "context_id": "b_6125c5000375",
            "observation": "absent"
          },
          {
            "context_id": "b_48b647235f33",
            "observation": "absent"
          },
          {
            "context_id": "b_42d9140f0598",
            "observation": "absent"
          }
        ],
        "uncertainty": "no other views confirm absence",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_2962b6747c5c",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "queries": [
          "black and white umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_2962b6747c5c_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "queries": [
          "red umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            646.9236450195312,
            822.9298706054688,
            966.198486328125,
            1159.1925048828125
          ]
        ],
        "box_scores": [
          0.28780749440193176
        ],
        "box_queries": [
          "red umbrella"
        ],
        "sam_predicted_iou": [
          0.984375
        ],
        "mask_path": "masks/b_2962b6747c5c_r2.png"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "blue umbrella",
        "queries": [
          "blue umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            646.8992309570312,
            822.8531494140625,
            966.4400024414062,
            1159.5849609375
          ]
        ],
        "box_scores": [
          0.2961234748363495
        ],
        "box_queries": [
          "blue umbrella"
        ],
        "sam_predicted_iou": [
          0.984375
        ],
        "mask_path": "masks/b_2962b6747c5c_r3.png"
      }
    ],
    "final_mask_path": "masks/b_2962b6747c5c.png",
    "routing_mask_path": "masks/b_2962b6747c5c_routing.png",
    "predicted_fraction": 0.037878212320974956
  }
}

0015 b_2d3a471eb3b2

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_2962b6747c5c
b_3d50f98d854f
b_42d9140f0598
b_66a730d6ef90
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_2962b6747c5c", "observation": "red umbrella is visible"},
      {"context_id": "b_3d50f98d854f", "observation": "red umbrella is visible"},
      {"context_id": "b_42d9140f0598", "observation": "red umbrella is visible"},
      {"context_id": "b_66a730d6ef90", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "black and white umbrella",
    "grounding_queries": ["black and white umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_2962b6747c5c", "observation": "black and white umbrella is visible"},
      {"context_id": "b_3d50f98d854f", "observation": "black and white umbrella is visible"},
      {"context_id": "b_42d9140f0598", "observation": "black and white umbrella is visible"},
      {"context_id": "b_66a730d6ef90", "observation": "black and white umbrella is visible"}
    ],
    "uncertainty": "black and white umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_2d3a471eb3b2",
    "target_sha256": "f014481f7f0078f0da0b52176b2bee217e86d6f0eadbe198bdd21db93b673616",
    "context_ids": [
      "b_2962b6747c5c",
      "b_3d50f98d854f",
      "b_42d9140f0598",
      "b_66a730d6ef90"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_2962b6747c5c",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_3d50f98d854f",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_42d9140f0598",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_66a730d6ef90",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "grounding_queries": [
          "black and white umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_2962b6747c5c",
            "observation": "black and white umbrella is visible"
          },
          {
            "context_id": "b_3d50f98d854f",
            "observation": "black and white umbrella is visible"
          },
          {
            "context_id": "b_42d9140f0598",
            "observation": "black and white umbrella is visible"
          },
          {
            "context_id": "b_66a730d6ef90",
            "observation": "black and white umbrella is visible"
          }
        ],
        "uncertainty": "black and white umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_2d3a471eb3b2",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_2d3a471eb3b2.png",
    "routing_mask_path": "masks/b_2d3a471eb3b2_routing.png",
    "predicted_fraction": 0.0
  }
}

0016 b_3d50f98d854f

train · IoU 0.068 · recall 0.069
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_42d9140f0598
b_66a730d6ef90
b_4dd28c1b3b60
b_2d3a471eb3b2
Dice 0.127Precision 0.738SafetyRecall 0.069BF@5 0.181Box coverage 0.079Overmask 0.009
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "mannequin",
    "grounding_queries": ["mannequin"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_42d9140f0598", "observation": "mannequin appears in the foreground"},
      {"context_id": "b_66a730d6ef90", "observation": "mannequin appears in the foreground"},
      {"context_id": "b_4dd28c1b3b60", "observation": "mannequin appears in the foreground"},
      {"context_id": "b_2d3a471eb3b2", "observation": "mannequin appears in the foreground"}
    ],
    "uncertainty": "mannequin is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.7, 0.3, 1.0, 0.8]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "mirror",
    "grounding_queries": ["mirror"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_4dd28c1b3b60", "observation": "mirror appears in the foreground"}
    ],
    "uncertainty": "mirror is only visible in one context image",
    "fallback_box_norm_xyxy": [0.6, 0.5, 0.9, 0.8]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_3d50f98d854f",
    "target_sha256": "b54ae4644bc87c5f8c5e614d9d4fd35577effb77828e182c45f19a0f901fdbbb",
    "context_ids": [
      "b_42d9140f0598",
      "b_66a730d6ef90",
      "b_4dd28c1b3b60",
      "b_2d3a471eb3b2"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mannequin",
        "grounding_queries": [
          "mannequin"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_42d9140f0598",
            "observation": "mannequin appears in the foreground"
          },
          {
            "context_id": "b_66a730d6ef90",
            "observation": "mannequin appears in the foreground"
          },
          {
            "context_id": "b_4dd28c1b3b60",
            "observation": "mannequin appears in the foreground"
          },
          {
            "context_id": "b_2d3a471eb3b2",
            "observation": "mannequin appears in the foreground"
          }
        ],
        "uncertainty": "mannequin is consistently visible across all context images",
        "fallback_box_norm_xyxy": [
          0.7,
          0.3,
          1.0,
          0.8
        ],
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "grounding_queries": [
          "mirror"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_4dd28c1b3b60",
            "observation": "mirror appears in the foreground"
          }
        ],
        "uncertainty": "mirror is only visible in one context image",
        "fallback_box_norm_xyxy": [
          0.6,
          0.5,
          0.9,
          0.8
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_3d50f98d854f",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "queries": [
          "mirror"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            707.57080078125,
            487.44549560546875,
            1005.8277587890625,
            673.9874877929688
          ],
          [
            939.605712890625,
            390.13330078125,
            1174.2569580078125,
            565.254150390625
          ]
        ],
        "box_scores": [
          0.3109780251979828,
          0.2871554493904114
        ],
        "box_queries": [
          "mirror",
          "mirror"
        ],
        "sam_predicted_iou": [
          0.9453125,
          0.94921875
        ],
        "mask_path": "masks/b_3d50f98d854f_r2.png"
      }
    ],
    "final_mask_path": "masks/b_3d50f98d854f.png",
    "routing_mask_path": "masks/b_3d50f98d854f_routing.png",
    "predicted_fraction": 0.02520717424844546
  }
}

0017 b_4dd28c1b3b60

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_d95203c68daa
b_3d50f98d854f
b_8ffaf4db41bf
b_17642d7e1164
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "mirror",
    "grounding_queries": ["mirror"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_d95203c68daa", "observation": "mirror is visible"},
      {"context_id": "b_3d50f98d854f", "observation": "mirror is visible"},
      {"context_id": "b_8ffaf4db41bf", "observation": "mirror is visible"},
      {"context_id": "b_17642d7e1164", "observation": "mirror is visible"}
    ],
    "uncertainty": "mirror is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "mirror",
    "grounding_queries": ["mirror"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_d95203c68daa", "observation": "mirror is visible"},
      {"context_id": "b_3d50f98d854f", "observation": "mirror is visible"},
      {"context_id": "b_8ffaf4db41bf", "observation": "mirror is visible"},
      {"context_id": "b_17642d7e1164", "observation": "mirror is visible"}
    ],
    "uncertainty": "mirror is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_4dd28c1b3b60",
    "target_sha256": "b1bcc7035d539193f181a7bb44e8eae918a065ce8bb46ebee596b324287323c8",
    "context_ids": [
      "b_d95203c68daa",
      "b_3d50f98d854f",
      "b_8ffaf4db41bf",
      "b_17642d7e1164"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "grounding_queries": [
          "mirror"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_d95203c68daa",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_3d50f98d854f",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_8ffaf4db41bf",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_17642d7e1164",
            "observation": "mirror is visible"
          }
        ],
        "uncertainty": "mirror is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "grounding_queries": [
          "mirror"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_d95203c68daa",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_3d50f98d854f",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_8ffaf4db41bf",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_17642d7e1164",
            "observation": "mirror is visible"
          }
        ],
        "uncertainty": "mirror is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_4dd28c1b3b60",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_4dd28c1b3b60.png",
    "routing_mask_path": "masks/b_4dd28c1b3b60_routing.png",
    "predicted_fraction": 0.0
  }
}

0018 b_d95203c68daa

train · IoU 0.608 · recall 0.685
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_4dd28c1b3b60
b_17642d7e1164
b_8ffaf4db41bf
b_42d3a74baab7
Dice 0.756Precision 0.844SafetyRecall 0.685BF@5 0.187Box coverage 0.861Overmask 0.017
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "mirror",
    "grounding_queries": ["mirror"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_4dd28c1b3b60", "observation": "mirror partially visible"},
      {"context_id": "b_17642d7e1164", "observation": "mirror fully visible"},
      {"context_id": "b_8ffaf4db41bf", "observation": "mirror partially visible"},
      {"context_id": "b_42d3a74baab7", "observation": "mirror fully visible"}
    ],
    "uncertainty": "mirror's position varies slightly across views",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "person holding mirror",
    "grounding_queries": ["person holding mirror"],
    "is_spatial": true,
    "confidence": 0.8,
    "context_evidence": [
      {"context_id": "b_4dd28c1b3b60", "observation": "person not visible"},
      {"context_id": "b_17642d7e1164", "observation": "person not visible"},
      {"context_id": "b_8ffaf4db41bf", "observation": "person not visible"},
      {"context_id": "b_42d3a74baab7", "observation": "person partially visible"}
    ],
    "uncertainty": "person's visibility is inconsistent",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_d95203c68daa",
    "target_sha256": "01bc9f3c3918db7763a12902c0f0e86339b42f1d5c6a592d7fe1465a67e76508",
    "context_ids": [
      "b_4dd28c1b3b60",
      "b_17642d7e1164",
      "b_8ffaf4db41bf",
      "b_42d3a74baab7"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "grounding_queries": [
          "mirror"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_4dd28c1b3b60",
            "observation": "mirror partially visible"
          },
          {
            "context_id": "b_17642d7e1164",
            "observation": "mirror fully visible"
          },
          {
            "context_id": "b_8ffaf4db41bf",
            "observation": "mirror partially visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "mirror fully visible"
          }
        ],
        "uncertainty": "mirror's position varies slightly across views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "person holding mirror",
        "grounding_queries": [
          "person holding mirror"
        ],
        "is_spatial": true,
        "confidence": 0.8,
        "context_evidence": [
          {
            "context_id": "b_4dd28c1b3b60",
            "observation": "person not visible"
          },
          {
            "context_id": "b_17642d7e1164",
            "observation": "person not visible"
          },
          {
            "context_id": "b_8ffaf4db41bf",
            "observation": "person not visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "person partially visible"
          }
        ],
        "uncertainty": "person's visibility is inconsistent",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_d95203c68daa",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "queries": [
          "mirror"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            1013.426025390625,
            278.7547912597656,
            1307.5675048828125,
            1166.7349853515625
          ],
          [
            639.6337280273438,
            490.9345703125,
            934.0805053710938,
            699.3328247070312
          ]
        ],
        "box_scores": [
          0.3015284538269043,
          0.26977986097335815
        ],
        "box_queries": [
          "mirror",
          "mirror"
        ],
        "sam_predicted_iou": [
          0.9296875,
          0.9296875
        ],
        "mask_path": "masks/b_d95203c68daa_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "person holding mirror",
        "queries": [
          "person holding mirror"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            1014.04736328125,
            278.1952209472656,
            1307.2713623046875,
            1166.5721435546875
          ]
        ],
        "box_scores": [
          0.314229816198349
        ],
        "box_queries": [
          "person holding mirror"
        ],
        "sam_predicted_iou": [
          0.9296875
        ],
        "mask_path": "masks/b_d95203c68daa_r2.png"
      }
    ],
    "final_mask_path": "masks/b_d95203c68daa.png",
    "routing_mask_path": "masks/b_d95203c68daa_routing.png",
    "predicted_fraction": 0.09696378200632724
  }
}

0019 b_c0cd1b9b9c0e

train · IoU 0.166 · recall 0.173
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_bc84988e0e0a
b_42d3a74baab7
b_c0a10b3aef20
b_17642d7e1164
Dice 0.285Precision 0.818SafetyRecall 0.173BF@5 0.066Box coverage 0.228Overmask 0.010
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_bc84988e0e0a", "observation": "red umbrella is visible"},
      {"context_id": "b_42d3a74baab7", "observation": "red umbrella is partially obscured"},
      {"context_id": "b_c0a10b3aef20", "observation": "red umbrella is fully obscured"},
      {"context_id": "b_17642d7e1164", "observation": "red umbrella is not visible"}
    ],
    "uncertainty": "red umbrella is only visible in one context",
    "fallback_box_norm_xyxy": [0.5, 0.4, 0.7, 0.5]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "scroll painting",
    "grounding_queries": ["scroll painting"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_bc84988e0e0a", "observation": "scroll painting is visible"},
      {"context_id": "b_42d3a74baab7", "observation": "scroll painting is partially obscured"},
      {"context_id": "b_c0a10b3aef20", "observation": "scroll painting is fully obscured"},
      {"context_id": "b_17642d7e1164", "observation": "scroll painting is not visible"}
    ],
    "uncertainty": "scroll painting is only visible in one context",
    "fallback_box_norm_xyxy": [0.6, 0.4, 0.8, 0.5]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_c0cd1b9b9c0e",
    "target_sha256": "fcb5d7fcc27dc2c89475f6d2923b6f666a06af139f14a1713be5fe487be7896b",
    "context_ids": [
      "b_bc84988e0e0a",
      "b_42d3a74baab7",
      "b_c0a10b3aef20",
      "b_17642d7e1164"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "red umbrella is partially obscured"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "red umbrella is fully obscured"
          },
          {
            "context_id": "b_17642d7e1164",
            "observation": "red umbrella is not visible"
          }
        ],
        "uncertainty": "red umbrella is only visible in one context",
        "fallback_box_norm_xyxy": [
          0.5,
          0.4,
          0.7,
          0.5
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "scroll painting",
        "grounding_queries": [
          "scroll painting"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "scroll painting is visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "scroll painting is partially obscured"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "scroll painting is fully obscured"
          },
          {
            "context_id": "b_17642d7e1164",
            "observation": "scroll painting is not visible"
          }
        ],
        "uncertainty": "scroll painting is only visible in one context",
        "fallback_box_norm_xyxy": [
          0.6,
          0.4,
          0.8,
          0.5
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_c0cd1b9b9c0e",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "queries": [
          "red umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            974.6791381835938,
            756.0464477539062,
            1211.8345947265625,
            1101.6326904296875
          ]
        ],
        "box_scores": [
          0.27580294013023376
        ],
        "box_queries": [
          "red umbrella"
        ],
        "sam_predicted_iou": [
          0.98046875
        ],
        "mask_path": "masks/b_c0cd1b9b9c0e_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "scroll painting",
        "queries": [
          "scroll painting"
        ],
        "source": "vlm_fallback",
        "boxes_xyxy": [
          [
            1054.8,
            467.20000000000005,
            1406.4,
            584.0
          ]
        ],
        "box_scores": [
          0.9
        ],
        "box_queries": [
          "vlm_fallback"
        ],
        "sam_predicted_iou": [
          0.8828125
        ],
        "mask_path": "masks/b_c0cd1b9b9c0e_r2.png"
      }
    ],
    "final_mask_path": "masks/b_c0cd1b9b9c0e.png",
    "routing_mask_path": "masks/b_c0cd1b9b9c0e_routing.png",
    "predicted_fraction": 0.04364003303878941
  }
}

0020 b_42d3a74baab7

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_17642d7e1164
b_c0cd1b9b9c0e
b_d95203c68daa
b_bc84988e0e0a
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "person holding mirror",
    "grounding_queries": ["person holding mirror"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_17642d7e1164", "observation": "mirror is visible"},
      {"context_id": "b_c0cd1b9b9c0e", "observation": "mirror is visible"},
      {"context_id": "b_d95203c68daa", "observation": "mirror is visible"},
      {"context_id": "b_bc84988e0e0a", "observation": "mirror is visible"}
    ],
    "uncertainty": "mirror is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_17642d7e1164", "observation": "red umbrella is visible"},
      {"context_id": "b_c0cd1b9b9c0e", "observation": "red umbrella is visible"},
      {"context_id": "b_d95203c68daa", "observation": "red umbrella is visible"},
      {"context_id": "b_bc84988e0e0a", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "color chart",
    "grounding_queries": ["color chart"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_17642d7e1164", "observation": "color chart is visible"},
      {"context_id": "b_c0cd1b9b9c0e", "observation": "color chart is visible"},
      {"context_id": "b_d95203c68daa", "observation": "color chart is visible"},
      {"context_id": "b_bc84988e0e0a", "observation": "color chart is visible"}
    ],
    "uncertainty": "color chart is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_42d3a74baab7",
    "target_sha256": "beecb1ddf9871d8ad3aff7649bc170553336bbfad9551fe94b82dfc99715c5ab",
    "context_ids": [
      "b_17642d7e1164",
      "b_c0cd1b9b9c0e",
      "b_d95203c68daa",
      "b_bc84988e0e0a"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "person holding mirror",
        "grounding_queries": [
          "person holding mirror"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_17642d7e1164",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_d95203c68daa",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "mirror is visible"
          }
        ],
        "uncertainty": "mirror is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_17642d7e1164",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_d95203c68daa",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "color chart",
        "grounding_queries": [
          "color chart"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_17642d7e1164",
            "observation": "color chart is visible"
          },
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "color chart is visible"
          },
          {
            "context_id": "b_d95203c68daa",
            "observation": "color chart is visible"
          },
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "color chart is visible"
          }
        ],
        "uncertainty": "color chart is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_42d3a74baab7",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_42d3a74baab7.png",
    "routing_mask_path": "masks/b_42d3a74baab7_routing.png",
    "predicted_fraction": 0.0
  }
}

0021 b_bc84988e0e0a

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_c0cd1b9b9c0e
b_45dcf0b2cf9e
b_42d3a74baab7
b_c0a10b3aef20
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "scroll with cranes",
    "grounding_queries": ["scroll with cranes"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_c0cd1b9b9c0e", "observation": "scroll with cranes is visible"},
      {"context_id": "b_45dcf0b2cf9e", "observation": "scroll with cranes is visible"},
      {"context_id": "b_42d3a74baab7", "observation": "scroll with cranes is visible"},
      {"context_id": "b_c0a10b3aef20", "observation": "scroll with cranes is visible"}
    ],
    "uncertainty": "scroll with cranes is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_c0cd1b9b9c0e", "observation": "red umbrella is visible"},
      {"context_id": "b_45dcf0b2cf9e", "observation": "red umbrella is visible"},
      {"context_id": "b_42d3a74baab7", "observation": "red umbrella is visible"},
      {"context_id": "b_c0a10b3aef20", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "color chart",
    "grounding_queries": ["color chart"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_c0cd1b9b9c0e", "observation": "color chart is visible"},
      {"context_id": "b_45dcf0b2cf9e", "observation": "color chart is visible"},
      {"context_id": "b_42d3a74baab7", "observation": "color chart is visible"},
      {"context_id": "b_c0a10b3aef20", "observation": "color chart is visible"}
    ],
    "uncertainty": "color chart is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_bc84988e0e0a",
    "target_sha256": "b2a22676fb275f1471a6a57d62769047f357b1dc24ac5d355f90ef4037ea737b",
    "context_ids": [
      "b_c0cd1b9b9c0e",
      "b_45dcf0b2cf9e",
      "b_42d3a74baab7",
      "b_c0a10b3aef20"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "scroll with cranes",
        "grounding_queries": [
          "scroll with cranes"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "scroll with cranes is visible"
          },
          {
            "context_id": "b_45dcf0b2cf9e",
            "observation": "scroll with cranes is visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "scroll with cranes is visible"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "scroll with cranes is visible"
          }
        ],
        "uncertainty": "scroll with cranes is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_45dcf0b2cf9e",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "color chart",
        "grounding_queries": [
          "color chart"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "color chart is visible"
          },
          {
            "context_id": "b_45dcf0b2cf9e",
            "observation": "color chart is visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "color chart is visible"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "color chart is visible"
          }
        ],
        "uncertainty": "color chart is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_bc84988e0e0a",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_bc84988e0e0a.png",
    "routing_mask_path": "masks/b_bc84988e0e0a_routing.png",
    "predicted_fraction": 0.0
  }
}

0022 b_45dcf0b2cf9e

train · IoU 0.005 · recall 0.005
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_bc84988e0e0a
b_c0cd1b9b9c0e
b_2f68fc214982
b_c0a10b3aef20
Dice 0.009Precision 0.052SafetyRecall 0.005BF@5 0.075Box coverage 0.032Overmask 0.020
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "black curtain",
    "grounding_queries": ["black curtain"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_c0a10b3aef20", "observation": "black curtain partially obscures the view"}
    ],
    "uncertainty": "curtain may be a reflection or specular effect",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_bc84988e0e0a", "observation": "red umbrella is visible"},
      {"context_id": "b_c0cd1b9b9c0e", "observation": "red umbrella is visible"},
      {"context_id": "b_2f68fc214982", "observation": "red umbrella is visible"},
      {"context_id": "b_c0a10b3aef20", "observation": "red umbrella is not visible"}
    ],
    "uncertainty": "umbrella may be a reflection or specular effect",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "white sign with red text",
    "grounding_queries": ["white sign with red text"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_bc84988e0e0a", "observation": "white sign with red text is visible"},
      {"context_id": "b_c0cd1b9b9c0e", "observation": "white sign with red text is visible"},
      {"context_id": "b_2f68fc214982", "observation": "white sign with red text is visible"},
      {"context_id": "b_c0a10b3aef20", "observation": "white sign with red text is not visible"}
    ],
    "uncertainty": "sign may be a reflection or specular effect",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_45dcf0b2cf9e",
    "target_sha256": "6094aaf918724214a99944de0f905327768498a637303147495f79f41fb064e1",
    "context_ids": [
      "b_bc84988e0e0a",
      "b_c0cd1b9b9c0e",
      "b_2f68fc214982",
      "b_c0a10b3aef20"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black curtain",
        "grounding_queries": [
          "black curtain"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "black curtain partially obscures the view"
          }
        ],
        "uncertainty": "curtain may be a reflection or specular effect",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_2f68fc214982",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "red umbrella is not visible"
          }
        ],
        "uncertainty": "umbrella may be a reflection or specular effect",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "white sign with red text",
        "grounding_queries": [
          "white sign with red text"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "white sign with red text is visible"
          },
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "white sign with red text is visible"
          },
          {
            "context_id": "b_2f68fc214982",
            "observation": "white sign with red text is visible"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "white sign with red text is not visible"
          }
        ],
        "uncertainty": "sign may be a reflection or specular effect",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_45dcf0b2cf9e",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black curtain",
        "queries": [
          "black curtain"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_45dcf0b2cf9e_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "queries": [
          "red umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            281.3736572265625,
            890.8401489257812,
            416.388916015625,
            1016.050048828125
          ],
          [
            1122.8323974609375,
            582.1143188476562,
            1406.656494140625,
            736.9505615234375
          ]
        ],
        "box_scores": [
          0.2556205093860626,
          0.2505408525466919
        ],
        "box_queries": [
          "red umbrella",
          "red umbrella"
        ],
        "sam_predicted_iou": [
          0.97265625,
          0.97265625
        ],
        "mask_path": "masks/b_45dcf0b2cf9e_r2.png"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "white sign with red text",
        "queries": [
          "white sign with red text"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            1122.9132080078125,
            581.886474609375,
            1406.521484375,
            736.3436889648438
          ]
        ],
        "box_scores": [
          0.32850027084350586
        ],
        "box_queries": [
          "white sign with red text"
        ],
        "sam_predicted_iou": [
          0.97265625
        ],
        "mask_path": "masks/b_45dcf0b2cf9e_r3.png"
      }
    ],
    "final_mask_path": "masks/b_45dcf0b2cf9e.png",
    "routing_mask_path": "masks/b_45dcf0b2cf9e_routing.png",
    "predicted_fraction": 0.017133027880374647
  }
}

0023 b_2f68fc214982

train · IoU 0.031 · recall 0.034
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_45dcf0b2cf9e
b_c0a10b3aef20
b_bc84988e0e0a
b_c0cd1b9b9c0e
Dice 0.061Precision 0.316SafetyRecall 0.034BF@5 0.090Box coverage 0.260Overmask 0.013
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "bear on floor",
    "grounding_queries": ["bear on floor"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_45dcf0b2cf9e", "observation": "bear partially obscured"},
      {"context_id": "b_c0a10b3aef20", "observation": "bear fully visible"},
      {"context_id": "b_bc84988e0e0a", "observation": "bear partially obscured"},
      {"context_id": "b_c0cd1b9b9c0e", "observation": "bear fully visible"}
    ],
    "uncertainty": "bear's position varies slightly across views",
    "fallback_box_norm_xyxy": [0.2, 0.5, 0.6, 0.8]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "umbrella on floor",
    "grounding_queries": ["umbrella on floor"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_45dcf0b2cf9e", "observation": "umbrella partially obscured"},
      {"context_id": "b_c0a10b3aef20", "observation": "umbrella fully visible"},
      {"context_id": "b_bc84988e0e0a", "observation": "umbrella partially obscured"},
      {"context_id": "b_c0cd1b9b9c0e", "observation": "umbrella fully visible"}
    ],
    "uncertainty": "umbrella's position varies slightly across views",
    "fallback_box_norm_xyxy": [0.5, 0.5, 0.7, 0.8]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "scroll on floor",
    "grounding_queries": ["scroll on floor"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_45dcf0b2cf9e", "observation": "scroll partially obscured"},
      {"context_id": "b_c0a10b3aef20", "observation": "scroll fully visible"},
      {"context_id": "b_bc84988e0e0a", "observation": "scroll partially obscured"},
      {"context_id": "b_c0cd1b9b9c0e", "observation": "scroll fully visible"}
    ],
    "uncertainty": "scroll's position varies slightly across views",
    "fallback_box_norm_xyxy": [0.7, 0.5, 0.9, 0.8]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_2f68fc214982",
    "target_sha256": "9aeb3f64365e0a88f38e6c4fe971124b99abdc8889daa561865c2178a8458f76",
    "context_ids": [
      "b_45dcf0b2cf9e",
      "b_c0a10b3aef20",
      "b_bc84988e0e0a",
      "b_c0cd1b9b9c0e"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "bear on floor",
        "grounding_queries": [
          "bear on floor"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_45dcf0b2cf9e",
            "observation": "bear partially obscured"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "bear fully visible"
          },
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "bear partially obscured"
          },
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "bear fully visible"
          }
        ],
        "uncertainty": "bear's position varies slightly across views",
        "fallback_box_norm_xyxy": [
          0.2,
          0.5,
          0.6,
          0.8
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "umbrella on floor",
        "grounding_queries": [
          "umbrella on floor"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_45dcf0b2cf9e",
            "observation": "umbrella partially obscured"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "umbrella fully visible"
          },
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "umbrella partially obscured"
          },
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "umbrella fully visible"
          }
        ],
        "uncertainty": "umbrella's position varies slightly across views",
        "fallback_box_norm_xyxy": [
          0.5,
          0.5,
          0.7,
          0.8
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "scroll on floor",
        "grounding_queries": [
          "scroll on floor"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_45dcf0b2cf9e",
            "observation": "scroll partially obscured"
          },
          {
            "context_id": "b_c0a10b3aef20",
            "observation": "scroll fully visible"
          },
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "scroll partially obscured"
          },
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "scroll fully visible"
          }
        ],
        "uncertainty": "scroll's position varies slightly across views",
        "fallback_box_norm_xyxy": [
          0.7,
          0.5,
          0.9,
          0.8
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_2f68fc214982",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "bear on floor",
        "queries": [
          "bear on floor"
        ],
        "source": "vlm_fallback",
        "boxes_xyxy": [
          [
            351.6,
            584.0,
            1054.8,
            934.4000000000001
          ]
        ],
        "box_scores": [
          0.9
        ],
        "box_queries": [
          "vlm_fallback"
        ],
        "sam_predicted_iou": [
          0.478515625
        ],
        "mask_path": "masks/b_2f68fc214982_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "umbrella on floor",
        "queries": [
          "umbrella on floor"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            138.65228271484375,
            864.3546752929688,
            303.8629150390625,
            1014.9258422851562
          ]
        ],
        "box_scores": [
          0.2737511098384857
        ],
        "box_queries": [
          "umbrella on floor"
        ],
        "sam_predicted_iou": [
          0.97265625
        ],
        "mask_path": "masks/b_2f68fc214982_r2.png"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "scroll on floor",
        "queries": [
          "scroll on floor"
        ],
        "source": "vlm_fallback",
        "boxes_xyxy": [
          [
            1230.6,
            584.0,
            1582.2,
            934.4000000000001
          ]
        ],
        "box_scores": [
          0.9
        ],
        "box_queries": [
          "vlm_fallback"
        ],
        "sam_predicted_iou": [
          0.83203125
        ],
        "mask_path": "masks/b_2f68fc214982_r3.png"
      }
    ],
    "final_mask_path": "masks/b_2f68fc214982.png",
    "routing_mask_path": "masks/b_2f68fc214982_routing.png",
    "predicted_fraction": 0.015903813486683186
  }
}

0024 b_c0a10b3aef20

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_c0cd1b9b9c0e
b_42d3a74baab7
b_bc84988e0e0a
b_2f68fc214982
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "black curtain",
    "grounding_queries": ["black curtain"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_c0cd1b9b9c0e", "observation": "black curtain partially visible"},
      {"context_id": "b_42d3a74baab7", "observation": "black curtain fully visible"},
      {"context_id": "b_bc84988e0e0a", "observation": "black curtain partially visible"},
      {"context_id": "b_2f68fc214982", "observation": "black curtain partially visible"}
    ],
    "uncertainty": "black curtain is consistently visible across all contexts.",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_c0cd1b9b9c0e", "observation": "red umbrella partially visible"},
      {"context_id": "b_42d3a74baab7", "observation": "red umbrella fully visible"},
      {"context_id": "b_bc84988e0e0a", "observation": "red umbrella partially visible"},
      {"context_id": "b_2f68fc214982", "observation": "red umbrella partially visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all contexts.",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "white signboard",
    "grounding_queries": ["white signboard"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_c0cd1b9b9c0e", "observation": "white signboard partially visible"},
      {"context_id": "b_42d3a74baab7", "observation": "white signboard fully visible"},
      {"context_id": "b_bc84988e0e0a", "observation": "white signboard partially visible"},
      {"context_id": "b_2f68fc214982", "observation": "white signboard partially visible"}
    ],
    "uncertainty": "white signboard is consistently visible across all contexts.",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_c0a10b3aef20",
    "target_sha256": "8d91c5d290a5064fe0a6fccead5e94c7d4909c3820c3252916712b856df52c85",
    "context_ids": [
      "b_c0cd1b9b9c0e",
      "b_42d3a74baab7",
      "b_bc84988e0e0a",
      "b_2f68fc214982"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black curtain",
        "grounding_queries": [
          "black curtain"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "black curtain partially visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "black curtain fully visible"
          },
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "black curtain partially visible"
          },
          {
            "context_id": "b_2f68fc214982",
            "observation": "black curtain partially visible"
          }
        ],
        "uncertainty": "black curtain is consistently visible across all contexts.",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "red umbrella partially visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "red umbrella fully visible"
          },
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "red umbrella partially visible"
          },
          {
            "context_id": "b_2f68fc214982",
            "observation": "red umbrella partially visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all contexts.",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "white signboard",
        "grounding_queries": [
          "white signboard"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "white signboard partially visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "white signboard fully visible"
          },
          {
            "context_id": "b_bc84988e0e0a",
            "observation": "white signboard partially visible"
          },
          {
            "context_id": "b_2f68fc214982",
            "observation": "white signboard partially visible"
          }
        ],
        "uncertainty": "white signboard is consistently visible across all contexts.",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_c0a10b3aef20",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_c0a10b3aef20.png",
    "routing_mask_path": "masks/b_c0a10b3aef20_routing.png",
    "predicted_fraction": 0.0
  }
}

0025 b_8ffaf4db41bf

train · IoU 0.139 · recall 0.139
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_d95203c68daa
b_4dd28c1b3b60
b_17642d7e1164
b_42d3a74baab7
Dice 0.245Precision 0.998SafetyRecall 0.139BF@5 0.082Box coverage 0.229Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "mirror",
    "grounding_queries": ["mirror"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_d95203c68daa", "observation": "mirror is visible"},
      {"context_id": "b_4dd28c1b3b60", "observation": "mirror is visible"},
      {"context_id": "b_17642d7e1164", "observation": "mirror is visible"},
      {"context_id": "b_42d3a74baab7", "observation": "mirror is partially visible"}
    ],
    "uncertainty": "mirror is partially visible in some views",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "person holding mirror",
    "grounding_queries": ["person holding mirror"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_d95203c68daa", "observation": "person holding mirror"},
      {"context_id": "b_4dd28c1b3b60", "observation": "person holding mirror"},
      {"context_id": "b_17642d7e1164", "observation": "person holding mirror"},
      {"context_id": "b_42d3a74baab7", "observation": "person holding mirror"}
    ],
    "uncertainty": "person holding mirror is partially visible",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_8ffaf4db41bf",
    "target_sha256": "215efed0325c4c5bfc4cab03cf298ffef068a601d17f729ebdc3803b1505db7d",
    "context_ids": [
      "b_d95203c68daa",
      "b_4dd28c1b3b60",
      "b_17642d7e1164",
      "b_42d3a74baab7"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "grounding_queries": [
          "mirror"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_d95203c68daa",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_4dd28c1b3b60",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_17642d7e1164",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "mirror is partially visible"
          }
        ],
        "uncertainty": "mirror is partially visible in some views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "person holding mirror",
        "grounding_queries": [
          "person holding mirror"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_d95203c68daa",
            "observation": "person holding mirror"
          },
          {
            "context_id": "b_4dd28c1b3b60",
            "observation": "person holding mirror"
          },
          {
            "context_id": "b_17642d7e1164",
            "observation": "person holding mirror"
          },
          {
            "context_id": "b_42d3a74baab7",
            "observation": "person holding mirror"
          }
        ],
        "uncertainty": "person holding mirror is partially visible",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_8ffaf4db41bf",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "queries": [
          "mirror"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            1185.1329345703125,
            284.9571533203125,
            1387.48193359375,
            921.6734619140625
          ]
        ],
        "box_scores": [
          0.2762335240840912
        ],
        "box_queries": [
          "mirror"
        ],
        "sam_predicted_iou": [
          0.93359375
        ],
        "mask_path": "masks/b_8ffaf4db41bf_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "person holding mirror",
        "queries": [
          "person holding mirror"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            1185.721435546875,
            284.9627380371094,
            1387.1141357421875,
            921.4432983398438
          ]
        ],
        "box_scores": [
          0.2745721936225891
        ],
        "box_queries": [
          "person holding mirror"
        ],
        "sam_predicted_iou": [
          0.93359375
        ],
        "mask_path": "masks/b_8ffaf4db41bf_r2.png"
      }
    ],
    "final_mask_path": "masks/b_8ffaf4db41bf.png",
    "routing_mask_path": "masks/b_8ffaf4db41bf_routing.png",
    "predicted_fraction": 0.03442384714884598
  }
}

0026 b_42d9140f0598

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_66a730d6ef90
b_3d50f98d854f
b_2d3a471eb3b2
b_4dd28c1b3b60
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.276Overmask 0.063
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "mannequin torso",
    "grounding_queries": ["mannequin torso"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_66a730d6ef90", "observation": "mannequin torso is visible"},
      {"context_id": "b_3d50f98d854f", "observation": "mannequin torso is visible"},
      {"context_id": "b_2d3a471eb3b2", "observation": "mannequin torso is visible"},
      {"context_id": "b_4dd28c1b3b60", "observation": "mannequin torso is partially visible"}
    ],
    "uncertainty": "mannequin torso is partially visible in some views",
    "fallback_box_norm_xyxy": [0.5, 0.0, 1.0, 0.5]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_66a730d6ef90", "observation": "red umbrella is visible"},
      {"context_id": "b_3d50f98d854f", "observation": "red umbrella is visible"},
      {"context_id": "b_2d3a471eb3b2", "observation": "red umbrella is visible"},
      {"context_id": "b_4dd28c1b3b60", "observation": "red umbrella is partially visible"}
    ],
    "uncertainty": "red umbrella is partially visible in some views",
    "fallback_box_norm_xyxy": [0.5, 0.5, 0.7, 0.7]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "color chart",
    "grounding_queries": ["color chart"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_66a730d6ef90", "observation": "color chart is visible"},
      {"context_id": "b_3d50f98d854f", "observation": "color chart is visible"},
      {"context_id": "b_2d3a471eb3b2", "observation": "color chart is visible"},
      {"context_id": "b_4dd28c1b3b60", "observation": "color chart is partially visible"}
    ],
    "uncertainty": "color chart is partially visible in some views",
    "fallback_box_norm_xyxy": [0.0, 0.5, 0.1, 0.6]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_42d9140f0598",
    "target_sha256": "267fc28ff59ccf3e906e9f6236c98d8f7ab103a5101b3228b4fed69b043b83f1",
    "context_ids": [
      "b_66a730d6ef90",
      "b_3d50f98d854f",
      "b_2d3a471eb3b2",
      "b_4dd28c1b3b60"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mannequin torso",
        "grounding_queries": [
          "mannequin torso"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_66a730d6ef90",
            "observation": "mannequin torso is visible"
          },
          {
            "context_id": "b_3d50f98d854f",
            "observation": "mannequin torso is visible"
          },
          {
            "context_id": "b_2d3a471eb3b2",
            "observation": "mannequin torso is visible"
          },
          {
            "context_id": "b_4dd28c1b3b60",
            "observation": "mannequin torso is partially visible"
          }
        ],
        "uncertainty": "mannequin torso is partially visible in some views",
        "fallback_box_norm_xyxy": [
          0.5,
          0.0,
          1.0,
          0.5
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_66a730d6ef90",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_3d50f98d854f",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_2d3a471eb3b2",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_4dd28c1b3b60",
            "observation": "red umbrella is partially visible"
          }
        ],
        "uncertainty": "red umbrella is partially visible in some views",
        "fallback_box_norm_xyxy": [
          0.5,
          0.5,
          0.7,
          0.7
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "color chart",
        "grounding_queries": [
          "color chart"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_66a730d6ef90",
            "observation": "color chart is visible"
          },
          {
            "context_id": "b_3d50f98d854f",
            "observation": "color chart is visible"
          },
          {
            "context_id": "b_2d3a471eb3b2",
            "observation": "color chart is visible"
          },
          {
            "context_id": "b_4dd28c1b3b60",
            "observation": "color chart is partially visible"
          }
        ],
        "uncertainty": "color chart is partially visible in some views",
        "fallback_box_norm_xyxy": [
          0.0,
          0.5,
          0.1,
          0.6
        ],
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_42d9140f0598",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mannequin torso",
        "queries": [
          "mannequin torso"
        ],
        "source": "vlm_fallback",
        "boxes_xyxy": [
          [
            879.0,
            0.0,
            1758.0,
            584.0
          ]
        ],
        "box_scores": [
          0.9
        ],
        "box_queries": [
          "vlm_fallback"
        ],
        "sam_predicted_iou": [
          0.890625
        ],
        "mask_path": "masks/b_42d9140f0598_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "queries": [
          "red umbrella"
        ],
        "source": "vlm_fallback",
        "boxes_xyxy": [
          [
            879.0,
            584.0,
            1230.6,
            817.5999999999999
          ]
        ],
        "box_scores": [
          0.9
        ],
        "box_queries": [
          "vlm_fallback"
        ],
        "sam_predicted_iou": [
          0.80078125
        ],
        "mask_path": "masks/b_42d9140f0598_r2.png"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "color chart",
        "queries": [
          "color chart"
        ],
        "source": "vlm_fallback",
        "boxes_xyxy": [
          [
            0.0,
            584.0,
            175.8,
            700.8
          ]
        ],
        "box_scores": [
          0.9
        ],
        "box_queries": [
          "vlm_fallback"
        ],
        "sam_predicted_iou": [
          0.86328125
        ],
        "mask_path": "masks/b_42d9140f0598_r3.png"
      }
    ],
    "final_mask_path": "masks/b_42d9140f0598.png",
    "routing_mask_path": "masks/b_42d9140f0598_routing.png",
    "predicted_fraction": 0.05020201193760032
  }
}

0027 b_48b647235f33

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_0a52973d1b8d
b_3fad5029fb0d
b_6125c5000375
b_2962b6747c5c
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "black and white umbrella",
    "grounding_queries": ["black and white umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_2962b6747c5c", "observation": "appears in one view"}
    ],
    "uncertainty": "insufficient evidence for persistence",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red cloth",
    "grounding_queries": ["red cloth"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_6125c5000375", "observation": "appears in one view"}
    ],
    "uncertainty": "insufficient evidence for persistence",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "green curtain",
    "grounding_queries": ["green curtain"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_3fad5029fb0d", "observation": "appears in one view"}
    ],
    "uncertainty": "insufficient evidence for persistence",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_48b647235f33",
    "target_sha256": "4478eb05fc8017524c6130aaf321ae48dcba6afa4f0b8a6f6bfcc8a7608b3d88",
    "context_ids": [
      "b_0a52973d1b8d",
      "b_3fad5029fb0d",
      "b_6125c5000375",
      "b_2962b6747c5c"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "grounding_queries": [
          "black and white umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_2962b6747c5c",
            "observation": "appears in one view"
          }
        ],
        "uncertainty": "insufficient evidence for persistence",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red cloth",
        "grounding_queries": [
          "red cloth"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_6125c5000375",
            "observation": "appears in one view"
          }
        ],
        "uncertainty": "insufficient evidence for persistence",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "green curtain",
        "grounding_queries": [
          "green curtain"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_3fad5029fb0d",
            "observation": "appears in one view"
          }
        ],
        "uncertainty": "insufficient evidence for persistence",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_48b647235f33",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "queries": [
          "black and white umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_48b647235f33_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red cloth",
        "queries": [
          "red cloth"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_48b647235f33_r2.png"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "green curtain",
        "queries": [
          "green curtain"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_48b647235f33_r3.png"
      }
    ],
    "final_mask_path": "masks/b_48b647235f33.png",
    "routing_mask_path": "masks/b_48b647235f33_routing.png",
    "predicted_fraction": 0.0
  }
}

0028 b_57cf54a3bbc0

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_0ed8205374dd
b_c7dd5015d023
b_31faa8f0af47
b_3fad5029fb0d
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "black curtain",
    "grounding_queries": ["black curtain"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_31faa8f0af47", "observation": "curtain partially visible"},
      {"context_id": "b_3fad5029fb0d", "observation": "curtain fully visible"}
    ],
    "uncertainty": "curtain may be partially obscured in some views",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_0ed8205374dd", "observation": "umbrella present"},
      {"context_id": "b_c7dd5015d023", "observation": "umbrella present"},
      {"context_id": "b_31faa8f0af47", "observation": "umbrella present"},
      {"context_id": "b_3fad5029fb0d", "observation": "umbrella present"}
    ],
    "uncertainty": "umbrella consistently visible",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_57cf54a3bbc0",
    "target_sha256": "afcf569e721329b5cb0850ec96c8f3bced66556ebadc4e3191fdb396f9f7cdcd",
    "context_ids": [
      "b_0ed8205374dd",
      "b_c7dd5015d023",
      "b_31faa8f0af47",
      "b_3fad5029fb0d"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black curtain",
        "grounding_queries": [
          "black curtain"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_31faa8f0af47",
            "observation": "curtain partially visible"
          },
          {
            "context_id": "b_3fad5029fb0d",
            "observation": "curtain fully visible"
          }
        ],
        "uncertainty": "curtain may be partially obscured in some views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_0ed8205374dd",
            "observation": "umbrella present"
          },
          {
            "context_id": "b_c7dd5015d023",
            "observation": "umbrella present"
          },
          {
            "context_id": "b_31faa8f0af47",
            "observation": "umbrella present"
          },
          {
            "context_id": "b_3fad5029fb0d",
            "observation": "umbrella present"
          }
        ],
        "uncertainty": "umbrella consistently visible",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_57cf54a3bbc0",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "black curtain",
        "queries": [
          "black curtain"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_57cf54a3bbc0_r1.png"
      }
    ],
    "final_mask_path": "masks/b_57cf54a3bbc0.png",
    "routing_mask_path": "masks/b_57cf54a3bbc0_routing.png",
    "predicted_fraction": 0.0
  }
}

0029 b_39a6a990391f

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_1ea940c8ac4e
b_31faa8f0af47
b_f9646cc9c954
b_0ed8205374dd
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "mirror",
    "grounding_queries": ["mirror"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_1ea940c8ac4e", "observation": "mirror is visible"},
      {"context_id": "b_31faa8f0af47", "observation": "mirror is visible"},
      {"context_id": "b_f9646cc9c954", "observation": "mirror is visible"},
      {"context_id": "b_0ed8205374dd", "observation": "mirror is visible"}
    ],
    "uncertainty": "mirror is consistently visible across all views",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "umbrella",
    "grounding_queries": ["umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_1ea940c8ac4e", "observation": "umbrella is visible"},
      {"context_id": "b_31faa8f0af47", "observation": "umbrella is visible"},
      {"context_id": "b_f9646cc9c954", "observation": "umbrella is visible"},
      {"context_id": "b_0ed8205374dd", "observation": "umbrella is visible"}
    ],
    "uncertainty": "umbrella is consistently visible across all views",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "pillow",
    "grounding_queries": ["pillow"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_1ea940c8ac4e", "observation": "pillow is visible"},
      {"context_id": "b_31faa8f0af47", "observation": "pillow is visible"},
      {"context_id": "b_f9646cc9c954", "observation": "pillow is visible"},
      {"context_id": "b_0ed8205374dd", "observation": "pillow is visible"}
    ],
    "uncertainty": "pillow is consistently visible across all views",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_39a6a990391f",
    "target_sha256": "f219da94ff53ac815426911c9fd887e919db470e608cea05b3894efd77048001",
    "context_ids": [
      "b_1ea940c8ac4e",
      "b_31faa8f0af47",
      "b_f9646cc9c954",
      "b_0ed8205374dd"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "grounding_queries": [
          "mirror"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_31faa8f0af47",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_f9646cc9c954",
            "observation": "mirror is visible"
          },
          {
            "context_id": "b_0ed8205374dd",
            "observation": "mirror is visible"
          }
        ],
        "uncertainty": "mirror is consistently visible across all views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "umbrella",
        "grounding_queries": [
          "umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_31faa8f0af47",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_f9646cc9c954",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_0ed8205374dd",
            "observation": "umbrella is visible"
          }
        ],
        "uncertainty": "umbrella is consistently visible across all views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "pillow",
        "grounding_queries": [
          "pillow"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "pillow is visible"
          },
          {
            "context_id": "b_31faa8f0af47",
            "observation": "pillow is visible"
          },
          {
            "context_id": "b_f9646cc9c954",
            "observation": "pillow is visible"
          },
          {
            "context_id": "b_0ed8205374dd",
            "observation": "pillow is visible"
          }
        ],
        "uncertainty": "pillow is consistently visible across all views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_39a6a990391f",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_39a6a990391f.png",
    "routing_mask_path": "masks/b_39a6a990391f_routing.png",
    "predicted_fraction": 0.0
  }
}

0030 b_e21902361190

train · IoU 0.000 · recall 0.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_6c74696a698a
b_c7168d6511cd
b_442bffc55440
b_f2d05c389498
Dice 0.000Precision 0.000SafetyRecall 0.000BF@5 0.000Box coverage 0.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_6c74696a698a", "observation": "red umbrella is visible"},
      {"context_id": "b_c7168d6511cd", "observation": "red umbrella is visible"},
      {"context_id": "b_442bffc55440", "observation": "red umbrella is visible"},
      {"context_id": "b_f2d05c389498", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [300, 270, 440, 380]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "white flower bouquet",
    "grounding_queries": ["white flower bouquet"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_6c74696a698a", "observation": "white flower bouquet is visible"},
      {"context_id": "b_c7168d6511cd", "observation": "white flower bouquet is visible"},
      {"context_id": "b_442bffc55440", "observation": "white flower bouquet is visible"},
      {"context_id": "b_f2d05c389498", "observation": "white flower bouquet is visible"}
    ],
    "uncertainty": "white flower bouquet is consistently visible across all context images",
    "fallback_box_norm_xyxy": [250, 100, 350, 200]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "green wall",
    "grounding_queries": ["green wall"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_6c74696a698a", "observation": "green wall is visible"},
      {"context_id": "b_c7168d6511cd", "observation": "green wall is visible"},
      {"context_id": "b_442bffc55440", "observation": "green wall is visible"},
      {"context_id": "b_f2d05c389498", "observation": "green wall is visible"}
    ],
    "uncertainty": "green wall is consistently visible across all context images",
    "fallback_box_norm_xyxy": [0, 0, 300, 300]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_e21902361190",
    "target_sha256": "7aad75333082d3e970f7577b36d73f84a6996115d0084e245bc54f56316fa30a",
    "context_ids": [
      "b_6c74696a698a",
      "b_c7168d6511cd",
      "b_442bffc55440",
      "b_f2d05c389498"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_6c74696a698a",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_c7168d6511cd",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_f2d05c389498",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "white flower bouquet",
        "grounding_queries": [
          "white flower bouquet"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_6c74696a698a",
            "observation": "white flower bouquet is visible"
          },
          {
            "context_id": "b_c7168d6511cd",
            "observation": "white flower bouquet is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "white flower bouquet is visible"
          },
          {
            "context_id": "b_f2d05c389498",
            "observation": "white flower bouquet is visible"
          }
        ],
        "uncertainty": "white flower bouquet is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "green wall",
        "grounding_queries": [
          "green wall"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_6c74696a698a",
            "observation": "green wall is visible"
          },
          {
            "context_id": "b_c7168d6511cd",
            "observation": "green wall is visible"
          },
          {
            "context_id": "b_442bffc55440",
            "observation": "green wall is visible"
          },
          {
            "context_id": "b_f2d05c389498",
            "observation": "green wall is visible"
          }
        ],
        "uncertainty": "green wall is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_e21902361190",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_e21902361190.png",
    "routing_mask_path": "masks/b_e21902361190_routing.png",
    "predicted_fraction": 0.0
  }
}

0031 b_442bffc55440

test · IoU 1.000 · recall 1.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_63ebbf87a7c1
b_a38e3ce498d5
b_e21902361190
b_6c74696a698a
Dice 1.000Precision 1.000SafetyRecall 1.000BF@5 1.000Box coverage 1.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_63ebbf87a7c1", "observation": "umbrella is visible"},
      {"context_id": "b_a38e3ce498d5", "observation": "umbrella is visible"},
      {"context_id": "b_e21902361190", "observation": "umbrella is visible"},
      {"context_id": "b_6c74696a698a", "observation": "umbrella is visible"}
    ],
    "uncertainty": "umbrella is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [294, 267, 400, 370]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "sign with Chinese characters",
    "grounding_queries": ["sign with Chinese characters"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_63ebbf87a7c1", "observation": "sign is visible"},
      {"context_id": "b_a38e3ce498d5", "observation": "sign is visible"},
      {"context_id": "b_e21902361190", "observation": "sign is visible"},
      {"context_id": "b_6c74696a698a", "observation": "sign is visible"}
    ],
    "uncertainty": "sign is consistently visible across all contexts",
    "fallback_box_norm_xyxy": [220, 400, 290, 476]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_442bffc55440",
    "target_sha256": "ea5e203e121b9e94bef7b63ba762139d4a7bfd98071db955ef86ed4c1923dbc0",
    "context_ids": [
      "b_63ebbf87a7c1",
      "b_a38e3ce498d5",
      "b_e21902361190",
      "b_6c74696a698a"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_e21902361190",
            "observation": "umbrella is visible"
          },
          {
            "context_id": "b_6c74696a698a",
            "observation": "umbrella is visible"
          }
        ],
        "uncertainty": "umbrella is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "sign with Chinese characters",
        "grounding_queries": [
          "sign with Chinese characters"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_63ebbf87a7c1",
            "observation": "sign is visible"
          },
          {
            "context_id": "b_a38e3ce498d5",
            "observation": "sign is visible"
          },
          {
            "context_id": "b_e21902361190",
            "observation": "sign is visible"
          },
          {
            "context_id": "b_6c74696a698a",
            "observation": "sign is visible"
          }
        ],
        "uncertainty": "sign is consistently visible across all contexts",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_442bffc55440",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_442bffc55440.png",
    "routing_mask_path": "masks/b_442bffc55440_routing.png",
    "predicted_fraction": 0.0
  }
}

0032 b_f2d05c389498

test · IoU 1.000 · recall 1.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_f9646cc9c954
b_e21902361190
b_c7168d6511cd
b_1ea940c8ac4e
Dice 1.000Precision 1.000SafetyRecall 1.000BF@5 1.000Box coverage 1.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_f9646cc9c954", "observation": "red umbrella is visible"},
      {"context_id": "b_e21902361190", "observation": "red umbrella is visible"},
      {"context_id": "b_c7168d6511cd", "observation": "red umbrella is visible"},
      {"context_id": "b_1ea940c8ac4e", "observation": "red umbrella is visible"}
    ],
    "uncertainty": "red umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [320, 224, 450, 345]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "black and white umbrella",
    "grounding_queries": ["black and white umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_f9646cc9c954", "observation": "black and white umbrella is visible"},
      {"context_id": "b_e21902361190", "observation": "black and white umbrella is visible"},
      {"context_id": "b_c7168d6511cd", "observation": "black and white umbrella is visible"},
      {"context_id": "b_1ea940c8ac4e", "observation": "black and white umbrella is visible"}
    ],
    "uncertainty": "black and white umbrella is consistently visible across all context images",
    "fallback_box_norm_xyxy": [140, 300, 260, 420]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_f2d05c389498",
    "target_sha256": "ce4ed46f67474b370e54193ad6ad1e8f2e0bf59b35a2c5e4e38dfe5a707dd6b6",
    "context_ids": [
      "b_f9646cc9c954",
      "b_e21902361190",
      "b_c7168d6511cd",
      "b_1ea940c8ac4e"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_f9646cc9c954",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_e21902361190",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_c7168d6511cd",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "red umbrella is visible"
          }
        ],
        "uncertainty": "red umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "black and white umbrella",
        "grounding_queries": [
          "black and white umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_f9646cc9c954",
            "observation": "black and white umbrella is visible"
          },
          {
            "context_id": "b_e21902361190",
            "observation": "black and white umbrella is visible"
          },
          {
            "context_id": "b_c7168d6511cd",
            "observation": "black and white umbrella is visible"
          },
          {
            "context_id": "b_1ea940c8ac4e",
            "observation": "black and white umbrella is visible"
          }
        ],
        "uncertainty": "black and white umbrella is consistently visible across all context images",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_f2d05c389498",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_f2d05c389498.png",
    "routing_mask_path": "masks/b_f2d05c389498_routing.png",
    "predicted_fraction": 0.0
  }
}

0033 b_0a52973d1b8d

test · IoU 1.000 · recall 1.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_48b647235f33
b_3fad5029fb0d
b_6125c5000375
b_c7dd5015d023
Dice 1.000Precision 1.000SafetyRecall 1.000BF@5 1.000Box coverage 1.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "red umbrella",
    "grounding_queries": ["red umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_48b647235f33", "observation": "red umbrella is visible"},
      {"context_id": "b_3fad5029fb0d", "observation": "red umbrella is visible"},
      {"context_id": "b_6125c5000375", "observation": "red umbrella is visible"},
      {"context_id": "b_c7dd5015d023", "observation": "red umbrella is not visible"}
    ],
    "uncertainty": "red umbrella is not visible in the last context image",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "green curtain",
    "grounding_queries": ["green curtain"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_48b647235f33", "observation": "green curtain is visible"},
      {"context_id": "b_3fad5029fb0d", "observation": "green curtain is visible"},
      {"context_id": "b_6125c5000375", "observation": "green curtain is visible"},
      {"context_id": "b_c7dd5015d023", "observation": "green curtain is not visible"}
    ],
    "uncertainty": "green curtain is not visible in the last context image",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r3",
    "cause": "transient_occluder",
    "semantic_description": "black chair",
    "grounding_queries": ["black chair"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_48b647235f33", "observation": "black chair is visible"},
      {"context_id": "b_3fad5029fb0d", "observation": "black chair is visible"},
      {"context_id": "b_6125c5000375", "observation": "black chair is visible"},
      {"context_id": "b_c7dd5015d023", "observation": "black chair is not visible"}
    ],
    "uncertainty": "black chair is not visible in the last context image",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_0a52973d1b8d",
    "target_sha256": "02d3286042e0c5392fa100b8c69e3a41868ea2bbd5a3cc9429cbb1b3789f65d7",
    "context_ids": [
      "b_48b647235f33",
      "b_3fad5029fb0d",
      "b_6125c5000375",
      "b_c7dd5015d023"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "grounding_queries": [
          "red umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_48b647235f33",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_3fad5029fb0d",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_6125c5000375",
            "observation": "red umbrella is visible"
          },
          {
            "context_id": "b_c7dd5015d023",
            "observation": "red umbrella is not visible"
          }
        ],
        "uncertainty": "red umbrella is not visible in the last context image",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "green curtain",
        "grounding_queries": [
          "green curtain"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_48b647235f33",
            "observation": "green curtain is visible"
          },
          {
            "context_id": "b_3fad5029fb0d",
            "observation": "green curtain is visible"
          },
          {
            "context_id": "b_6125c5000375",
            "observation": "green curtain is visible"
          },
          {
            "context_id": "b_c7dd5015d023",
            "observation": "green curtain is not visible"
          }
        ],
        "uncertainty": "green curtain is not visible in the last context image",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "black chair",
        "grounding_queries": [
          "black chair"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_48b647235f33",
            "observation": "black chair is visible"
          },
          {
            "context_id": "b_3fad5029fb0d",
            "observation": "black chair is visible"
          },
          {
            "context_id": "b_6125c5000375",
            "observation": "black chair is visible"
          },
          {
            "context_id": "b_c7dd5015d023",
            "observation": "black chair is not visible"
          }
        ],
        "uncertainty": "black chair is not visible in the last context image",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_0a52973d1b8d",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "red umbrella",
        "queries": [
          "red umbrella"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_0a52973d1b8d_r1.png"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "green curtain",
        "queries": [
          "green curtain"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_0a52973d1b8d_r2.png"
      },
      {
        "region_id": "r3",
        "cause": "transient_occluder",
        "semantic_description": "black chair",
        "queries": [
          "black chair"
        ],
        "source": "grounding",
        "boxes_xyxy": [],
        "box_scores": [],
        "box_queries": [],
        "sam_predicted_iou": [],
        "mask_path": "masks/b_0a52973d1b8d_r3.png"
      }
    ],
    "final_mask_path": "masks/b_0a52973d1b8d.png",
    "routing_mask_path": "masks/b_0a52973d1b8d_routing.png",
    "predicted_fraction": 0.0
  }
}

0034 b_66a730d6ef90

test · IoU 1.000 · recall 1.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_42d9140f0598
b_3d50f98d854f
b_2d3a471eb3b2
b_2962b6747c5c
Dice 1.000Precision 1.000SafetyRecall 1.000BF@5 1.000Box coverage 1.000Overmask 0.000
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "mannequin",
    "grounding_queries": ["mannequin"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_42d9140f0598", "observation": "mannequin partially obscures the teddy bear"},
      {"context_id": "b_3d50f98d854f", "observation": "mannequin partially obscures the teddy bear"},
      {"context_id": "b_2d3a471eb3b2", "observation": "mannequin partially obscures the teddy bear"},
      {"context_id": "b_2962b6747c5c", "observation": "mannequin partially obscures the teddy bear"}
    ],
    "uncertainty": "mannequin is partially visible in all context images, indicating it is not a spatial nuisance.",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  },
  {
    "region_id": "r2",
    "cause": "transient_occluder",
    "semantic_description": "umbrella",
    "grounding_queries": ["umbrella"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_42d9140f0598", "observation": "umbrella partially obscures the teddy bear"},
      {"context_id": "b_3d50f98d854f", "observation": "umbrella partially obscures the teddy bear"},
      {"context_id": "b_2d3a471eb3b2", "observation": "umbrella partially obscures the teddy bear"},
      {"context_id": "b_2962b6747c5c", "observation": "umbrella partially obscures the teddy bear"}
    ],
    "uncertainty": "umbrella is partially visible in all context images, indicating it is not a spatial nuisance.",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_66a730d6ef90",
    "target_sha256": "2caa6d5ab56862347537483d461fa637f435a1d2209c4e4106f78cab764e5592",
    "context_ids": [
      "b_42d9140f0598",
      "b_3d50f98d854f",
      "b_2d3a471eb3b2",
      "b_2962b6747c5c"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mannequin",
        "grounding_queries": [
          "mannequin"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_42d9140f0598",
            "observation": "mannequin partially obscures the teddy bear"
          },
          {
            "context_id": "b_3d50f98d854f",
            "observation": "mannequin partially obscures the teddy bear"
          },
          {
            "context_id": "b_2d3a471eb3b2",
            "observation": "mannequin partially obscures the teddy bear"
          },
          {
            "context_id": "b_2962b6747c5c",
            "observation": "mannequin partially obscures the teddy bear"
          }
        ],
        "uncertainty": "mannequin is partially visible in all context images, indicating it is not a spatial nuisance.",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      },
      {
        "region_id": "r2",
        "cause": "transient_occluder",
        "semantic_description": "umbrella",
        "grounding_queries": [
          "umbrella"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_42d9140f0598",
            "observation": "umbrella partially obscures the teddy bear"
          },
          {
            "context_id": "b_3d50f98d854f",
            "observation": "umbrella partially obscures the teddy bear"
          },
          {
            "context_id": "b_2d3a471eb3b2",
            "observation": "umbrella partially obscures the teddy bear"
          },
          {
            "context_id": "b_2962b6747c5c",
            "observation": "umbrella partially obscures the teddy bear"
          }
        ],
        "uncertainty": "umbrella is partially visible in all context images, indicating it is not a spatial nuisance.",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": false,
        "consistency_filter": "rejected_explicitly_persistent"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_66a730d6ef90",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [],
    "final_mask_path": "masks/b_66a730d6ef90.png",
    "routing_mask_path": "masks/b_66a730d6ef90_routing.png",
    "predicted_fraction": 0.0
  }
}

0035 b_17642d7e1164

test · IoU 0.000 · recall 1.000
blind target
SAM prediction (red)
paired-reference pseudo-GT (green, evaluation only)
error: TP green / FP red / FN blue
reference, opened only after inference lock
4 pose-nearest context images
b_42d3a74baab7
b_d95203c68daa
b_c0cd1b9b9c0e
b_8ffaf4db41bf
Dice 0.000Precision 0.000SafetyRecall 1.000BF@5 0.000Box coverage 1.000Overmask 0.015
VLM raw response
```json
[
  {
    "region_id": "r1",
    "cause": "transient_occluder",
    "semantic_description": "mirror",
    "grounding_queries": ["mirror"],
    "is_spatial": true,
    "confidence": 0.9,
    "context_evidence": [
      {"context_id": "b_42d3a74baab7", "observation": "mirror partially obscures the teddy bear"},
      {"context_id": "b_d95203c68daa", "observation": "mirror fully obscures the teddy bear"},
      {"context_id": "b_c0cd1b9b9c0e", "observation": "mirror partially obscures the teddy bear"},
      {"context_id": "b_8ffaf4db41bf", "observation": "mirror fully obscures the teddy bear"}
    ],
    "uncertainty": "mirror's position varies across views",
    "fallback_box_norm_xyxy": [0.0, 0.0, 1.0, 1.0]
  }
]
```
Structured VLM + grounding/SAM
{
  "vlm": {
    "schema_version": "diag.region.v1",
    "sample_id": "b_17642d7e1164",
    "target_sha256": "ee1fbf84bece876f54c8a63e2d4bbff0086d2ee90c0ed535b19e4040d6381f54",
    "context_ids": [
      "b_42d3a74baab7",
      "b_d95203c68daa",
      "b_c0cd1b9b9c0e",
      "b_8ffaf4db41bf"
    ],
    "model": {
      "provider": "local",
      "name": "qwen2.5-vl-7b-instruct",
      "version": "local_snapshot",
      "quantization": "nf4"
    },
    "prompt_sha256": "15497acb4f0a5266ff625a2745e66a38aa75c362c8f6e3fe7b4bfb2bc5003a36",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "grounding_queries": [
          "mirror"
        ],
        "is_spatial": true,
        "confidence": 0.9,
        "context_evidence": [
          {
            "context_id": "b_42d3a74baab7",
            "observation": "mirror partially obscures the teddy bear"
          },
          {
            "context_id": "b_d95203c68daa",
            "observation": "mirror fully obscures the teddy bear"
          },
          {
            "context_id": "b_c0cd1b9b9c0e",
            "observation": "mirror partially obscures the teddy bear"
          },
          {
            "context_id": "b_8ffaf4db41bf",
            "observation": "mirror fully obscures the teddy bear"
          }
        ],
        "uncertainty": "mirror's position varies across views",
        "fallback_box_norm_xyxy": null,
        "routing_eligible": true,
        "consistency_filter": "passed"
      }
    ],
    "global_causes": [],
    "parse_error": null,
    "format_repair": "wrapped_top_level_region_list",
    "validation_errors": []
  },
  "grounding": {
    "schema_version": "diag.ground_sam.v1",
    "sample_id": "b_17642d7e1164",
    "thresholds": {
      "vlm_confidence": 0.35,
      "box": 0.25,
      "text": 0.2,
      "nms": 0.5
    },
    "grounding_model": "/local/scratch-2/fy277/temp/diag3dgs/models/grounding-dino-tiny",
    "sam_checkpoint": "/local/scratch-2/fy277/temp/diag3dgs/models/sam2.1/sam2.1_hiera_small.pt",
    "sam_checkpoint_sha256": "6d1aa6f30de5c92224f8172114de081d104bbd23dd9dc5c58996f0cad5dc4d38",
    "regions": [
      {
        "region_id": "r1",
        "cause": "transient_occluder",
        "semantic_description": "mirror",
        "queries": [
          "mirror"
        ],
        "source": "grounding",
        "boxes_xyxy": [
          [
            592.1898193359375,
            452.15679931640625,
            883.3391723632812,
            680.7554931640625
          ]
        ],
        "box_scores": [
          0.3116818070411682
        ],
        "box_queries": [
          "mirror"
        ],
        "sam_predicted_iou": [
          0.9296875
        ],
        "mask_path": "masks/b_17642d7e1164_r1.png"
      }
    ],
    "final_mask_path": "masks/b_17642d7e1164.png",
    "routing_mask_path": "masks/b_17642d7e1164_routing.png",
    "predicted_fraction": 0.015087096950145713
  }
}