Video Generation on VideoJAM (test)
0.5782Aesthetic QualityLocalDPO
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| LocalDPOModel Backbone=CogvideoX-5B, Optimization Strategy=Ours2026.01 | 0.5782 | 0.6727 | 0.2523 | 21.03 | 0.5707 | 1.8785 | 0.319 | 5.4451 | 7.6424 | |
| LocalDPOModel Backbone=Wan 2.1-1.3B, Optimization Strategy=Ours2026.01 | 0.5698 | 0.6467 | 0.2533 | 20.92 | 0.6667 | 1.7033 | 0.2366 | 5.545 | 7.4849 | |
| SFTModel Backbone=Wan 2.1-1.3B, Optimization Strategy=SFT2026.01 | 0.5675 | 0.6003 | 0.2494 | 20.81 | 0.6302 | 1.3571 | 0.1555 | 5.5195 | 7.0321 | |
| SFTModel Backbone=CogvideoX-5B, Optimization Strategy=SFT2026.01 | 0.5635 | 0.6166 | 0.2445 | 20.99 | 0.5485 | 1.7151 | 0.2771 | 5.4379 | 7.4301 | |
| BaselineModel Backbone=CogvideoX-5B, Optimization Strategy=Baseline2026.01 | 0.5631 | 0.6135 | 0.2421 | 21 | 0.4805 | 1.7597 | 0.2987 | 5.4428 | 7.5012 | |
| BaselineModel Backbone=Wan 2.1-1.3B, Optimization Strategy=Baseline2026.01 | 0.5623 | 0.6021 | 0.2499 | 20.82 | 0.6292 | 1.3637 | 0.1613 | 5.6295 | 7.1545 | |
| DenseDPOModel Backbone=Wan 2.1-1.3B, Optimization Strategy=DenseDPO2026.01 | 0.5622 | 0.6021 | 0.2501 | 20.82 | 0.6342 | 1.3657 | 0.1387 | 5.6156 | 7.12 | |
| DenseDPOModel Backbone=CogvideoX-5B, Optimization Strategy=DenseDPO2026.01 | 0.5614 | 0.6171 | 0.2424 | 20.98 | 0.5188 | 1.7947 | 0.264 | 5.3634 | 7.422 | |
| Vanilla DPOModel Backbone=Wan 2.1-1.3B, Optimization Strategy=Vanilla DPO2026.01 | 0.5611 | 0.6042 | 0.2503 | 20.83 | 0.6496 | 1.3646 | 0.1357 | 5.5545 | 7.0548 | |
| LocalDPOModel Backbone=CogvideoX-2B, Optimization Strategy=Ours2026.01 | 0.5604 | 0.7001 | 0.2543 | 20.97 | 0.7036 | 1.8207 | 0.3134 | 5.4054 | 7.5397 | |
| SFTModel Backbone=CogvideoX-2B, Optimization Strategy=SFT2026.01 | 0.5567 | 0.6382 | 0.2471 | 21.04 | 0.691 | 1.7966 | 0.33 | 5.3368 | 7.4635 | |
| Vanilla DPOModel Backbone=CogvideoX-5B, Optimization Strategy=Vanilla DPO2026.01 | 0.5553 | 0.6148 | 0.2403 | 20.94 | 0.4996 | 1.7056 | 0.2785 | 5.3235 | 7.3076 | |
| DenseDPOModel Backbone=CogvideoX-2B, Optimization Strategy=DenseDPO2026.01 | 0.5521 | 0.6334 | 0.2448 | 20.96 | 0.6501 | 1.8147 | 0.3568 | 5.3559 | 7.5214 | |
| BaselineModel Backbone=CogvideoX-2B, Optimization Strategy=Baseline2026.01 | 0.5494 | 0.6327 | 0.2445 | 20.88 | 0.6407 | 1.7707 | 0.3849 | 5.314 | 7.4696 | |
| Vanilla DPOModel Backbone=CogvideoX-2B, Optimization Strategy=Vanilla DPO2026.01 | 0.5482 | 0.631 | 0.2443 | 20.96 | 0.6358 | 1.8198 | 0.3446 | 5.3515 | 7.516 |