text-conditioned human interaction generation on InterX (test)
43.5R-Precision (Top 1)Ground Truth
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Ground Truth2025.12 | 43.5 | 62.8 | 73.6 | 0.002 | 3.574 | 8.947 | — | |
| ARMFlowType=Online, CFG strength=1.2, Granularity=4-frame2025.12 | 42 | 60.6 | 71.1 | 0.042 | 3.728 | 8.939 | 1.203 | |
| InterMask2024.10 | 40.3 | 59.5 | 70.5 | 0.399 | 3.705 | 9.046 | 2.261 | |
| ReGenNetType=Online, Causal Masking=true2025.12 | 40.2 | 58.4 | 69 | 0.071 | 3.843 | 9.011 | 0.738 | |
| CAMDMType=Online, Context Window=10 frames2025.12 | 31.2 | 48 | 58.7 | 0.429 | 4.468 | 8.467 | 1.46 | |
| InterFormerType=Online2025.12 | 30.1 | 46.9 | 57.1 | 0.304 | 4.604 | 8.579 | 0.289 | |
| InterGen2024.10 | 20.7 | 33.5 | 42.9 | 5.207 | 9.58 | 7.788 | 3.686 | |
| MDM2024.10 | 20.3 | 32.9 | 42.6 | 23.701 | 9.548 | 5.856 | 3.49 | |
| T2M2024.10 | 18.4 | 29.8 | 39.6 | 5.481 | 9.576 | 2.771 | 2.761 | |
| ComMDM2024.10 | 9 | 16.5 | 23.6 | 29.266 | 6.87 | 4.734 | 0.771 |