松田研究室
Run SummaryRun ID: hotdog-brushcompat-opacdecay-30k
Workspace: splat / commit 5501a1e
← Run Index
Run summary · P1.M5 multi-scene parity (低 wallclock + 低 splats 版)

hotdog-brushcompat-opacdecay-30k

P1.D multi-scene re-chain: hotdog brushcompat + opacity_decay 30k

p1-dmulti-scenebrush-compatpremultipliedopacity-decayphase-d-multihotdog
partial pass
01

設定

dataset
/Users/otkrickey/dev/3dgs-workspace/datasets/nerf_synthetic/hotdog
gt_convention
Premultiplied
iterations
30,000
seed
42
capacity
1,000,000 splats
sh_degree
3
loss
L1Ssim
lambda
0.200
ssim
window=7 sigma=1
backend.backward
Simd
backend.loss_path
Gpu
02

結果

Val PSNR
37.33dB
Wallclock
23m 39s
Final splats
206,620
Final loss
2.601626e-3
03

3D ビューワー

学習結果 PLY を .splat 形式に変換し、antimatter15/splat の WebGL ビューワーで対話表示。

04

推移

Loss curve — 反復ごとの training loss
1e-2.71e-2.11e-1.51e-0.91e-0.36,00012k18k24k30kthis runloss
Splat count — refine の発火と最終構成
-4,89350,656106,204161,752217,3016,00012k18k24k30kthis runsplats
05

中間ログ抜粋

iter本 run loss
14.3290e-1
1,0001.8472e-2
2,0006.8285e-3
3,0004.3147e-3
4,0003.8332e-3
5,0003.5775e-3
6,0003.3416e-3
7,0003.2864e-3
8,0003.1690e-3
9,0003.1206e-3
10,0003.1333e-3
11,0003.0362e-3
12,0003.0276e-3
13,0002.9908e-3
14,0002.9450e-3
15,0002.9651e-3
16,0002.8382e-3
17,0002.8350e-3
18,0002.7661e-3
19,0002.7255e-3
20,0002.6714e-3
21,0002.6888e-3
22,0002.6185e-3
23,0002.6183e-3
24,0002.6631e-3
25,0002.6317e-3
26,0002.5914e-3
27,0002.6635e-3
28,0002.6030e-3
29,0002.5633e-3
30,0002.6016e-3
06

関連 finding

この run を related_runs に含む finding (Positive / Negative どちらも)。

© 2026 大竹律輝 — 松田研究室