DiagSplat · all-30 transient experiment

Raw vs Wild vs pseudo-mask v1 vs refined-mask 3DGS

All four models use the same 30 transient training views and five clean held-out views. Pseudo-mask v1 is the older run stored internally as oracle_route=typed; it uses the first-round pseudo masks, not clean oracle images. The refined-mask model uses the final GPT boundary-refined masks. The trajectory below is a genuine novel-view interpolation, not a video made from the five test images.

Masked vs raw PSNR+4.75 dB
Masked vs Wild PSNR+1.47 dB
Shared trajectory238 frames30 fps · 800×532

Shared novel-view trajectory

Left to right: Raw 3DGS · WildGaussians · Pseudo-mask v1 · Refined-mask 3DGS. The two mask routes use the exact same 238 trajectory poses (trajectory JSON SHA-256 matched).

Five clean held-out test views

These are the same five clean targets used for the quantitative evaluation.

Held-out view 1

Clean GT
Raw 3DGS
WildGaussians
Pseudo-mask v1
Refined-mask 3DGS

Held-out view 2

Clean GT
Raw 3DGS
WildGaussians
Pseudo-mask v1
Refined-mask 3DGS

Held-out view 3

Clean GT
Raw 3DGS
WildGaussians
Pseudo-mask v1
Refined-mask 3DGS

Held-out view 4

Clean GT
Raw 3DGS
WildGaussians
Pseudo-mask v1
Refined-mask 3DGS

Held-out view 5

Clean GT
Raw 3DGS
WildGaussians
Pseudo-mask v1
Refined-mask 3DGS

Shared test-view metrics

All three rows use the same Graphdeco evaluator and VGG-LPIPS implementation.

MethodPSNR ↑SSIM ↑LPIPS ↓
Raw 3DGS22.2950.7800.293
WildGaussians25.5780.8680.284
Pseudo-mask v126.7270.8150.249
Refined-mask 3DGS27.0440.8150.249

Summary: refined masks improve raw 3DGS by 4.75 dB PSNR, +0.035 SSIM, and 0.044 LPIPS. Versus WildGaussians, refined-mask 3DGS has +1.47 dB PSNR and +0.035 LPIPS, while WildGaussians has the higher SSIM by 0.053.

What changed from pseudo-mask v1?

The boundary-refined masks improve mean PSNR by +0.318 dB, while SSIM changes by -0.000064 and LPIPS by +0.000283. Held-out view 4 changes only from 25.758 dB to 25.785 dB. This isolates its remaining artifact as a supervision/geometry limitation rather than a mask-boundary error.

Synchronized individual videos

Raw 3DGS

WildGaussians

Pseudo-mask v1

Refined-mask 3DGS

Trajectory construction

Natural cubic translation and smooth SO(3) rotation interpolation through the 30 ordered cameras, followed by a reverse traversal for a continuous ping-pong loop.

Six trajectory checkpoints