松田研究室
Run SummaryRun ID: ship-brushcompat-30k
Workspace: splat / commit dc6a529
← Run Index
Run summary · P1.M5 multi-scene parity

ship-brushcompat-30k

P1.B+F Stage 2 multi-scene: ship brushcompat 30k

p1-b-fmulti-scenebrush-compatconvention-bridgepremultipliedstage-2-multiship
partial pass
01

設定

dataset
/Users/otkrickey/dev/3dgs-workspace/datasets/nerf_synthetic/ship
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
30.37dB
Wallclock
1h 4m 21s
Final splats
897,259
Final loss
8.622854e-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-2.21e-1.71e-1.11e-0.61e-0.16,00012k18k24k30kthis runloss
Splat count — refine の発火と最終構成
-39,396205,919451,233696,547941,8626,00012k18k24k30kthis runsplats
05

中間ログ抜粋

iter本 run loss
15.8242e-1
1,0004.2935e-2
2,0002.3085e-2
3,0001.6940e-2
4,0001.4758e-2
5,0001.3630e-2
6,0001.3104e-2
7,0001.2501e-2
8,0001.2131e-2
9,0001.1832e-2
10,0001.1374e-2
11,0001.1167e-2
12,0001.1075e-2
13,0001.0830e-2
14,0001.0727e-2
15,0001.0608e-2
16,0009.8128e-3
17,0009.7839e-3
18,0009.5663e-3
19,0009.4466e-3
20,0009.2521e-3
21,0009.2100e-3
22,0009.0719e-3
23,0009.0380e-3
24,0008.9821e-3
25,0008.9414e-3
26,0008.8927e-3
27,0008.7972e-3
28,0008.6969e-3
29,0008.6969e-3
30,0008.6229e-3
© 2026 大竹律輝 — 松田研究室