Checkerboard wrasse · w12_fish2
Where it comes from
3D and 4D model
Previous version (v1) vs v2
gs_final, 4000 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 | 4000 | 12000 | |
| gaussians | 16,881 | 40,247 | x2.38 |
| training time (optimisation) | 20 min | 72 min | |
| held-out frames (within 0.3 s of a training frame) | 40 (7) | 40 (7) | |
| LPIPS on the fish (lower = closer to the photo) | 0.2214 | 0.1660 | -0.0555 |
| PSNR full crop (dB) | 22.87 | 22.80 | -0.07 |
| PSNR on the fish mask (dB) | 18.59 | 18.58 | -0.01 |
| sharpness ratio render / photo, grain-robust (1 = as sharp as the photo) | 0.199 | 0.522 | x2.62 |
| sharpness ratio, raw | 0.057 | 0.355 | x6.27 |
| sharp match (photo detail reproduced in place) | 0.069 | 0.204 | +0.136 |
| IoU with the SAM mask | 0.892 | 0.881 | -0.011 |
Verdict: better. LPIPS -0.0555, grain-robust sharpness x2.62 (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: Labridae 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 | 12:00 – 22:00 |
| visits | 10: 13:23-13:25, 13:43-13:46, 13:53-14:05, 14:46-15:03, 15:39-15:47, 15:51-16:10, 16:12-16:19, 16:50-16:55, 21:45-21:52, 21:56-22:00 |
| frames (10 fps) detected / kept | 713 / 709 |
| valid poses (10 fps) | 650 |
| frames with background plate | 655 |
| median size | 147.0 px |
| 30 fps frames / valid poses | 677 / 569 |
| method | 4D GS, fins + bending + tail (30 fps) |
| model shown | v2 (gs_v2, quality experiment rollout) |
| iterations | 12000 |
| gaussians | 40247 |
| training time | 72 min |
| trainer validation PSNR full / masked | 25.1 / 17.6 dB |
| trainer validation IoU vs SAM mask | 0.872 |
| near-fish band PSNR (fins) | 27.2 dB |
| production status | done |
| visual note (curation) | checkerboard wrasse, Halichoeres hortulanus 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