01
設定
- dataset
- /Users/otkrickey/dev/3dgs-workspace/datasets/nerf_synthetic/ship
- 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
15.04dB
Wallclock
23m 58s
Final splats
107,450
Final loss
3.437925e-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 | 3.7113e-1 |
| 500 | 9.4673e-2 |
| 1,000 | 9.9741e-2 |
| 1,500 | 1.0450e-1 |
| 2,000 | 6.4857e-2 |
| 2,500 | 5.2411e-2 |
| 3,000 | 5.0366e-2 |
| 3,500 | 4.7058e-2 |
| 4,000 | 4.5601e-2 |
| 4,500 | 4.3959e-2 |
| 5,000 | 4.3361e-2 |
| 5,500 | 4.2261e-2 |
| 6,000 | 4.1208e-2 |
| 6,500 | 4.0520e-2 |
| 7,000 | 3.9732e-2 |
| 7,500 | 3.9410e-2 |
| 8,000 | 3.9231e-2 |
| 8,500 | 4.3678e-2 |
| 9,000 | 3.9050e-2 |
| 9,500 | 3.8556e-2 |
| 10,000 | 3.8118e-2 |
| 10,500 | 3.7552e-2 |
| 11,000 | 3.7487e-2 |
| 11,500 | 3.7436e-2 |
| 12,000 | 3.8216e-2 |
| 12,500 | 3.7419e-2 |
| 13,000 | 3.6973e-2 |
| 13,500 | 3.7056e-2 |
| 14,000 | 3.6650e-2 |
| 14,500 | 3.6935e-2 |
| 15,000 | 3.6094e-2 |
| 15,500 | 3.6144e-2 |
| 16,000 | 3.6099e-2 |
| 16,500 | 3.5845e-2 |
| 17,000 | 3.5724e-2 |
| 17,500 | 3.6166e-2 |
| 18,000 | 3.5456e-2 |
| 18,500 | 3.6497e-2 |
| 19,000 | 3.5309e-2 |
| 19,500 | 3.5039e-2 |
| 20,000 | 3.5778e-2 |
| 20,500 | 3.5331e-2 |
| 21,000 | 3.5422e-2 |
| 21,500 | 3.5460e-2 |
| 22,000 | 3.4993e-2 |
| 22,500 | 3.5051e-2 |
| 23,000 | 3.5025e-2 |
| 23,500 | 3.5042e-2 |
| 24,000 | 3.5094e-2 |
| 24,500 | 3.5487e-2 |
| 25,000 | 3.4993e-2 |
| 25,500 | 3.6087e-2 |
| 26,000 | 3.4778e-2 |
| 26,500 | 3.4733e-2 |
| 27,000 | 3.4556e-2 |
| 27,500 | 3.4928e-2 |
| 28,000 | 3.4592e-2 |
| 28,500 | 3.4923e-2 |
| 29,000 | 3.4292e-2 |
| 29,500 | 3.4372e-2 |
| 30,000 | 3.4379e-2 |
07
関連 finding
この run を related_runs に含む finding (Positive / Negative どちらも)。