Orange-lined triggerfish · fish2
Where it comes from
3D and 4D model
Previous version (v1) vs v2
gs_final, 10000 iterations) with 12000 iterations and a gaussian cap of 100,000 — the only change of the quality experiment that passed its review. Other changes (frame selection, water and motion-blur models, white balance, lossless crops) did not. The two turntables play in sync (same orbit and timing).
| metric | v1 gs_final | v2 gs_v2 | change |
|---|---|---|---|
| iterations | 10000 | 12000 | |
| gaussians | 26,908 | 29,482 | x1.10 |
| training time (optimisation) | 62 min | 80 min | |
| held-out frames (within 0.3 s of a training frame) | 16 (16) | 16 (16) | |
| LPIPS on the fish (lower = closer to the photo) | 0.1694 | 0.1672 | -0.0022 |
| PSNR full crop (dB) | 24.14 | 24.16 | +0.02 |
| PSNR on the fish mask (dB) | 21.31 | 21.32 | +0.01 |
| sharpness ratio render / photo, grain-robust (1 = as sharp as the photo) | 0.321 | 0.347 | x1.08 |
| sharpness ratio, raw | 0.119 | 0.154 | x1.29 |
| sharp match (photo detail reproduced in place) | 0.279 | 0.298 | +0.019 |
| IoU with the SAM mask | 0.897 | 0.899 | +0.002 |
Verdict: same (within noise). LPIPS -0.0022, grain-robust sharpness x1.08 (sharper). Differences below 0.0034 LPIPS or x1.12 sharpness are within the run-to-run noise of training.
Held-out frames: fixed per fish (work/quality_exp/<fish>/heldout.json), evaluated identically for both models (test-time fit of pose, colour and bending with the gaussians frozen). Both models were trained on all frames of the same data set, so held-out frames near a training frame favour both equally; this is a like-for-like comparison of the two recipes, not a generalisation test.
Poster
Species
Family: Balistidae 100%, Serranidae 0%, Pseudochromidae 0% · agreement across crops 100% · 64 crops · bioclip-2.5 (hf-hub:imageomics/bioclip-2.5-vith14, open_clip, zero-shot)
Data and reconstruction
| window | 48:00 – 58:00 |
| visits | 5: 48:02-48:12, 48:23-48:29, 49:41-49:47, 51:16-51:20, 52:58-53:09 |
| frames (10 fps) detected / kept | 389 / 359 |
| valid poses (10 fps) | 312 |
| frames with background plate | 316 |
| median size | 241.0 px |
| 30 fps frames / valid poses | 1146 / 881 |
| method | 4D GS, fins + bending + tail (30 fps) |
| model shown | v2 (gs_v2, quality experiment rollout) |
| iterations | 12000 |
| gaussians | 29482 |
| training time | 80 min |
| trainer validation PSNR full / masked | 25.5 / 20.4 dB |
| trainer validation IoU vs SAM mask | 0.890 |
| near-fish band PSNR (fins) | 25.9 dB |
| production status | done; gs_final = copy of fish2_hf/gs_fins_4d (10000 iters, not retrained) |
| visual note (curation) | Balistapus undulatus (orange-lined triggerfish), likely |
Diagnostics
Downloads
point_cloud.ply (canonical 3D gaussians) · poses_refined.json · model.pt (incl. per-frame 4D params) · species.json · previous version v1: gs_final point_cloud.ply