Iterative VLM transient mask bootstrapping · v0

0.302proposal GT coverage
0.395SAM oracle IoU
0.000final mask IoU
0.000final mask recall
6.7%clean-frame FPR
1accepted proposals

Clean test reconstruction

10k unmasked control: PSNR 17.6992, SSIM 0.6532, LPIPS 0.5528
10k predicted-mask Round 1: PSNR 17.6962, SSIM 0.6532, LPIPS 0.5528

Reference-derived pseudo-GT appears only below this locked evaluation section.

0001 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[
  {
    "proposal_id": "round_00__view_0001__proposal_01",
    "decision": "transient",
    "confidence": 0.9,
    "selected_sam_candidate": "A",
    "object_description": "umbrella",
    "evidence_for": "The umbrella appears only in the target frame and is not present in the context frames.",
    "evidence_against": "The umbrella is not present in the static reconstruction or the RGB residual proposal visualization.",
    "needs_more_evidence": false,
    "frame_id": "0001",
    "raw_response": "```json\n{\n  \"decision\": \"transient\",\n  \"confidence\": 0.9,\n  \"selected_sam_candidate\": \"A\",\n  \"object_description\": \"umbrella\",\n  \"evidence_for\": \"The umbrella appears only in the target frame and is not present in the context frames.\",\n  \"evidence_against\": \"The umbrella is not present in the static reconstruction or the RGB residual proposal visualization.\",\n  \"needs_more_evidence\": false\n}\n```"
  }
]

0002 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.101 · SAM oracle 0.311

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0003 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0004 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.211 · SAM oracle 0.086

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0005 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0006 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.207 · SAM oracle 0.031

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0007 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0008 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.672 · SAM oracle 0.539

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0009 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0010 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.022 · SAM oracle 0.025

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0011 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0012 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.026

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0013 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0014 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.061 · SAM oracle 0.049

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0015 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0016 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.396 · SAM oracle 0.707

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0017 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0018 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.457 · SAM oracle 0.745

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0019 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0020 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.155 · SAM oracle 0.268

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0021 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0022 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.510 · SAM oracle 0.800

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0023 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0024 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.580 · SAM oracle 0.687

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0025 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0026 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.293 · SAM oracle 0.606

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0027 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0028 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.605 · SAM oracle 0.625

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0029 · clean

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.000 · SAM oracle 0.000

obs
render
residual
proposal
pred
gt
VLM decisions
[]

0030 · dynamic

IoU 0.000 · precision 0.000 · recall 0.000 · proposal coverage 0.257 · SAM oracle 0.422

obs
render
residual
proposal
pred
gt
VLM decisions
[]