← Run Index
Run summary · P1.M5 multi-scene parity
hotdog-brushcompat-30k
P1.B+F Stage 2 multi-scene: hotdog brushcompat 30k
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
37.14dB
Wallclock
38m 47s
Final splats
485,357
Final loss
2.431413e-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
Splat count — refine の発火と最終構成
05
中間ログ抜粋
| iter | 本 run loss |
|---|---|
| 1 | 4.3290e-1 |
| 1,000 | 1.8067e-2 |
| 2,000 | 7.6700e-3 |
| 3,000 | 4.7227e-3 |
| 4,000 | 4.0408e-3 |
| 5,000 | 3.7741e-3 |
| 6,000 | 3.5498e-3 |
| 7,000 | 3.4118e-3 |
| 8,000 | 3.2272e-3 |
| 9,000 | 3.1831e-3 |
| 10,000 | 3.1137e-3 |
| 11,000 | 3.0569e-3 |
| 12,000 | 3.0043e-3 |
| 13,000 | 2.9197e-3 |
| 14,000 | 2.8950e-3 |
| 15,000 | 2.8975e-3 |
| 16,000 | 2.7864e-3 |
| 17,000 | 2.7162e-3 |
| 18,000 | 2.7344e-3 |
| 19,000 | 2.6708e-3 |
| 20,000 | 2.6285e-3 |
| 21,000 | 2.5861e-3 |
| 22,000 | 2.5755e-3 |
| 23,000 | 2.5639e-3 |
| 24,000 | 2.5686e-3 |
| 25,000 | 2.5231e-3 |
| 26,000 | 2.4978e-3 |
| 27,000 | 2.4602e-3 |
| 28,000 | 2.4958e-3 |
| 29,000 | 2.4270e-3 |
| 30,000 | 2.4314e-3 |