← Run Index
Run summary · Phase 5
lego-cap50000-30k
E.6 capacity scaling: lego sh3 30k capacity=50000 (モバイル制約下の最適 capacity 探索)
partial pass
01
設定
- dataset
- /Users/otkrickey/dev/3dgs-workspace/datasets/nerf_synthetic/lego
- iterations
- 30,000
- seed
- 42
- capacity
- 50,000 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
25.28dB
Wallclock
21m 35s
Final splats
82,550
Final loss
2.486539e-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 | 1.1788e-1 |
| 1,000 | 1.0172e-1 |
| 1,500 | 8.2831e-2 |
| 2,000 | 5.6547e-2 |
| 2,500 | 4.7380e-2 |
| 3,000 | 4.2377e-2 |
| 3,500 | 3.8194e-2 |
| 4,000 | 3.5604e-2 |
| 4,500 | 3.3975e-2 |
| 5,000 | 3.2788e-2 |
| 5,500 | 3.1240e-2 |
| 6,000 | 3.1216e-2 |
| 6,500 | 3.1305e-2 |
| 7,000 | 2.9975e-2 |
| 7,500 | 2.9560e-2 |
| 8,000 | 2.9178e-2 |
| 8,500 | 2.7547e-2 |
| 9,000 | 2.7808e-2 |
| 9,500 | 2.8003e-2 |
| 10,000 | 2.7428e-2 |
| 10,500 | 2.7372e-2 |
| 11,000 | 2.6767e-2 |
| 11,500 | 2.7370e-2 |
| 12,000 | 2.6786e-2 |
| 12,500 | 2.6688e-2 |
| 13,000 | 2.7351e-2 |
| 13,500 | 2.6868e-2 |
| 14,000 | 2.6835e-2 |
| 14,500 | 2.5408e-2 |
| 15,000 | 2.5863e-2 |
| 15,500 | 2.5370e-2 |
| 16,000 | 2.5552e-2 |
| 16,500 | 2.6619e-2 |
| 17,000 | 2.6309e-2 |
| 17,500 | 2.5028e-2 |
| 18,000 | 2.6044e-2 |
| 18,500 | 2.5671e-2 |
| 19,000 | 2.7010e-2 |
| 19,500 | 2.6190e-2 |
| 20,000 | 2.4707e-2 |
| 20,500 | 2.6359e-2 |
| 21,000 | 2.4479e-2 |
| 21,500 | 2.5107e-2 |
| 22,000 | 2.4900e-2 |
| 22,500 | 2.4290e-2 |
| 23,000 | 2.5105e-2 |
| 23,500 | 2.4422e-2 |
| 24,000 | 2.4121e-2 |
| 24,500 | 2.4962e-2 |
| 25,000 | 2.6211e-2 |
| 25,500 | 2.5487e-2 |
| 26,000 | 2.4952e-2 |
| 26,500 | 2.4880e-2 |
| 27,000 | 2.5006e-2 |
| 27,500 | 2.6133e-2 |
| 28,000 | 2.4535e-2 |
| 28,500 | 2.5435e-2 |
| 29,000 | 2.5902e-2 |
| 29,500 | 2.4310e-2 |
| 30,000 | 2.4865e-2 |
07
関連 finding
この run を related_runs に含む finding (Positive / Negative どちらも)。