← Run Index
Run summary · Phase 5
lego-cap200000-30k
E.6 capacity scaling: lego sh3 30k capacity=200000 (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
- 200,000 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
24.60dB
Wallclock
21m 42s
Final splats
85,021
Final loss
2.479345e-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 | 9.8646e-2 |
| 1,500 | 8.0682e-2 |
| 2,000 | 5.5204e-2 |
| 2,500 | 4.4785e-2 |
| 3,000 | 3.7747e-2 |
| 3,500 | 3.6713e-2 |
| 4,000 | 3.3532e-2 |
| 4,500 | 3.2041e-2 |
| 5,000 | 3.0150e-2 |
| 5,500 | 3.0239e-2 |
| 6,000 | 2.9949e-2 |
| 6,500 | 2.9865e-2 |
| 7,000 | 2.9755e-2 |
| 7,500 | 2.7583e-2 |
| 8,000 | 2.7774e-2 |
| 8,500 | 2.7300e-2 |
| 9,000 | 2.6863e-2 |
| 9,500 | 2.6358e-2 |
| 10,000 | 2.6512e-2 |
| 10,500 | 2.5722e-2 |
| 11,000 | 2.6631e-2 |
| 11,500 | 2.5611e-2 |
| 12,000 | 2.6265e-2 |
| 12,500 | 2.5182e-2 |
| 13,000 | 2.5521e-2 |
| 13,500 | 2.5641e-2 |
| 14,000 | 2.6685e-2 |
| 14,500 | 2.4980e-2 |
| 15,000 | 2.3840e-2 |
| 15,500 | 2.5142e-2 |
| 16,000 | 2.5438e-2 |
| 16,500 | 2.5113e-2 |
| 17,000 | 2.4677e-2 |
| 17,500 | 2.6052e-2 |
| 18,000 | 2.5227e-2 |
| 18,500 | 2.4382e-2 |
| 19,000 | 2.4877e-2 |
| 19,500 | 2.4671e-2 |
| 20,000 | 2.4825e-2 |
| 20,500 | 2.4048e-2 |
| 21,000 | 2.4430e-2 |
| 21,500 | 2.6158e-2 |
| 22,000 | 2.3815e-2 |
| 22,500 | 2.4220e-2 |
| 23,000 | 2.4617e-2 |
| 23,500 | 2.4636e-2 |
| 24,000 | 2.3807e-2 |
| 24,500 | 2.4171e-2 |
| 25,000 | 2.3782e-2 |
| 25,500 | 2.3542e-2 |
| 26,000 | 2.3921e-2 |
| 26,500 | 2.3970e-2 |
| 27,000 | 2.4141e-2 |
| 27,500 | 2.4695e-2 |
| 28,000 | 2.3269e-2 |
| 28,500 | 2.4292e-2 |
| 29,000 | 2.3096e-2 |
| 29,500 | 2.3070e-2 |
| 30,000 | 2.4793e-2 |
07
関連 finding
この run を related_runs に含む finding (Positive / Negative どちらも)。