Checkerboard wrasse · w02_fish1

Halichoeres hortulanus · Labridae · 100% · high reconstructed model: good

Reviewer: checker pattern and yellow tail

Where it comes from

Original video (longest visit, 07:54-08:04), the selected fish boxed

Tracking: real frame + crop + 3D render at the estimated pose, species overlay

3D and 4D model

loading 3D gaussians…

Interactive canonical 3D gaussians (drag to orbit, scroll to zoom). This is the rest pose shared by every frame; the 4D part is the per-frame bending / tail motion shown on the right.

4D: canonical model replaying the learned body bending and tail beat

3D turntable of the canonical model

Previous version (v1) vs v2

v2 = the same 30 fps data and trainer as v1 (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.

v1 (previous version): gs_final, 4000 iterations

v2: gs_v2, 12000 iterations

The two turntables play in sync (same orbit and timing).

metric	v1 gs_final	v2 gs_v2	change
iterations	4000	12000
gaussians	18,790	40,251	x2.14
training time (optimisation)	19 min	73 min
held-out frames (within 0.3 s of a training frame)	21 (10)	21 (10)
LPIPS on the fish (lower = closer to the photo)	0.1882	0.1589	-0.0293
PSNR full crop (dB)	24.62	24.69	+0.07
PSNR on the fish mask (dB)	19.26	19.44	+0.18
sharpness ratio render / photo, grain-robust (1 = as sharp as the photo)	0.321	0.468	x1.46
sharpness ratio, raw	0.137	0.281	x2.05
sharp match (photo detail reproduced in place)	0.145	0.251	+0.105
IoU with the SAM mask	0.901	0.897	-0.004

Verdict: better. LPIPS -0.0293, grain-robust sharpness x1.46 (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

Checkerboard wrasse Halichoeres hortulanus Labridae

100.0%

Zigzag wrasse Halichoeres scapularis Labridae

0.0%

Rusty parrotfish Scarus ferrugineus Scaridae

0.0%

Chevron butterflyfish Chaetodon trifascialis Chaetodontidae

0.0%

Red Sea longnose filefish Oxymonacanthus halli Monacanthidae

0.0%

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	02:00 – 12:00
visits	12: 02:46-02:50, 04:16-04:19, 05:12-05:17, 05:41-05:44, 07:55-08:03, 08:13-08:19, 08:33-08:36, 08:40-08:42, 08:45-08:52, 08:56-08:59, 09:19-09:21, 09:27-09:30
frames (10 fps) detected / kept	491 / 461
valid poses (10 fps)	426
frames with background plate	420
median size	239.0 px
30 fps frames / valid poses	724 / 592
method	4D GS, fins + bending + tail (30 fps)
model shown	v2 (gs_v2, quality experiment rollout)
iterations	12000
gaussians	40251
training time	73 min
trainer validation PSNR full / masked	25.1 / 18.6 dB
trainer validation IoU vs SAM mask	0.884
near-fish band PSNR (fins)	27.0 dB
production status	done
visual note (curation)	checkerboard wrasse, Halichoeres hortulanus likely

Diagnostics

Held-out frames: real crop vs 4D render (composited over the plate)

Estimated pose: fitted ellipsoid + axes on crops

Frame cleaning: identity / occlusion over time

Frames down-weighted / dropped by the robust trainer

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

← all fish