← Run Index
Run summary · P1 Phase G.5 final Pareto integration
lego-c3-tile-counting-12500
C3 (per-tile counting sort) GPU gate — full 12.5k quality guard run vs lego-bench-calib-checkpoints baseline (PSNR 36.226 dB / 360s)
partial pass
01
設定
- dataset
- /Users/otkrickey/dev/3dgs-workspace/datasets/nerf_synthetic/lego
- gt_convention
- Premultiplied
- iterations
- 12,500
- 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
- backend.emit_pairs
- Original
- backend.forward
- F32
- backend.dispatch
- PerStage
- backend.radix_scan
- Cpu
- backend.binning
- Resident
- backend.tile_sort
- TileCounting
- backend.target_cache
- true
- backend.shader_fast_math
- true
- os_build
- 26.5.2 (25F84)
02
結果
Val PSNR
36.19dB
Wallclock
5m 29s
Final splats
405,293
Final loss
5.627568e-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 | 5.9385e-1 |
| 1,000 | 5.4564e-2 |
| 2,000 | 1.4257e-2 |
| 3,000 | 8.3704e-3 |
| 4,000 | 7.4275e-3 |
| 5,000 | 6.5542e-3 |
| 6,000 | 6.2588e-3 |
| 7,000 | 6.0423e-3 |
| 8,000 | 5.8889e-3 |
| 9,000 | 5.7828e-3 |
| 10,000 | 5.6984e-3 |
| 11,000 | 5.6995e-3 |
| 12,000 | 5.6015e-3 |
| 12,500 | 5.6276e-3 |
06
関連 finding
この run を related_runs に含む finding (Positive / Negative どちらも)。