Human Preference Alignment on HPD v2
0.379HPS-v2.1BranchGRPO
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BranchGRPO2025.11 | 0.379 | — | 6.571 | 0.362 | 3.138 | |
| NeighborGRPOp=22025.11 | 0.379 | — | 6.503 | 0.368 | 3.132 | |
| NeighborGRPOp=0.82025.11 | 0.372 | — | 6.626 | 0.371 | 3.166 | |
| MixGRPO2025.11 | 0.371 | — | 6.365 | 0.367 | 3.125 | |
| TreeGRPOIter. Time (s)=79.22025.12 | 0.364 | 1.3426 | 6.4237 | 0.383 | — | |
| DanceGRPO2025.11 | 0.356 | — | 6.4 | 0.366 | 3.114 | |
| MixGRPOIter. Time (s)=1522025.12 | 0.3521 | 1.2056 | 6.0488 | 0.3812 | — | |
| DanceGRPOIter. Time (s)=1842025.12 | 0.3485 | 1.393 | 6.3224 | 0.3862 | — | |
| DDPOIter. Time (s)=178.22025.12 | 0.2748 | 1.0061 | 5.85 | 0.3884 | — | |
| SD3.5-M2025.12 | 0.2725 | 0.887 | 5.9519 | 0.3996 | — |