← Run Index
Run summary · Phase 5
lego-iter25000
E.5 iter scaling: lego sh3 25000 iter (収益逓減 plateau 探索)
partial pass
01
設定
- dataset
- /Users/otkrickey/dev/3dgs-workspace/datasets/nerf_synthetic/lego
- iterations
- 25,000
- 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
24.62dB
Wallclock
18m 19s
Final splats
84,086
Final loss
2.757549e-2
03
描画結果
学習後の eval pipeline から取得した代表視点。
00
01
0204
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。
05
推移
Loss curve — 反復ごとの training loss
Splat count — refine の発火と最終構成
06
中間ログ抜粋
| iter | 本 run loss |
|---|---|
| 1 | 3.4534e-1 |
| 500 | 1.1786e-1 |
| 1,000 | 1.0038e-1 |
| 1,500 | 8.0207e-2 |
| 2,000 | 5.2913e-2 |
| 2,500 | 4.3864e-2 |
| 3,000 | 3.9849e-2 |
| 3,500 | 3.6988e-2 |
| 4,000 | 3.4899e-2 |
| 4,500 | 3.3665e-2 |
| 5,000 | 3.1507e-2 |
| 5,500 | 3.1350e-2 |
| 6,000 | 2.9803e-2 |
| 6,500 | 2.9393e-2 |
| 7,000 | 2.9367e-2 |
| 7,500 | 2.8458e-2 |
| 8,000 | 2.9042e-2 |
| 8,500 | 2.9107e-2 |
| 9,000 | 2.8087e-2 |
| 9,500 | 2.8341e-2 |
| 10,000 | 2.8250e-2 |
| 10,500 | 2.6642e-2 |
| 11,000 | 2.6171e-2 |
| 11,500 | 2.6230e-2 |
| 12,000 | 2.7363e-2 |
| 12,500 | 2.5689e-2 |
| 13,000 | 2.6502e-2 |
| 13,500 | 2.7467e-2 |
| 14,000 | 2.5514e-2 |
| 14,500 | 2.5396e-2 |
| 15,000 | 2.6900e-2 |
| 15,500 | 2.5960e-2 |
| 16,000 | 2.5241e-2 |
| 16,500 | 2.5642e-2 |
| 17,000 | 2.5243e-2 |
| 17,500 | 2.5774e-2 |
| 18,000 | 2.6167e-2 |
| 18,500 | 2.5867e-2 |
| 19,000 | 2.4574e-2 |
| 19,500 | 2.4987e-2 |
| 20,000 | 2.5229e-2 |
| 20,500 | 2.5499e-2 |
| 21,000 | 2.4671e-2 |
| 21,500 | 2.7151e-2 |
| 22,000 | 2.4997e-2 |
| 22,500 | 2.5117e-2 |
| 23,000 | 2.3848e-2 |
| 23,500 | 2.7594e-2 |
| 24,000 | 2.3998e-2 |
| 24,500 | 2.5588e-2 |
| 25,000 | 2.7575e-2 |
07
関連 finding
この run を related_runs に含む finding (Positive / Negative どちらも)。