Dyadic Gesture Generation on Seamless Interaction Dataset (test)
33.85Diversity (Static)GT Random
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GT RandomType=Ground Truth Random2026.02 | 33.85 | — | 14.94 | 3.74 | 2.05 | |
| GTType=Ground Truth2026.02 | 28.42 | — | — | — | 1.97 | |
| DyaDiTConditioning=Full2026.02 | 27.46 | 7.71 | 6.4 | 1.37 | 1.38 | |
| DyaDiTConditioning=w/o self2026.02 | 26.43 | 7.61 | 6.64 | 1.48 | 1.25 | |
| DyaDiTConditioning=w/o ORCA2026.02 | 23.57 | 7.25 | 7.32 | 1.79 | 1.24 | |
| DyaDiTConditioning=Random2026.02 | 21.94 | 7.64 | 8.24 | 1.53 | 1.43 | |
| DyaDiTConditioning=Uncond2026.02 | 21.65 | 7.63 | 7.4 | 1.63 | 1.16 | |
| DyaDiTConditioning=MD contin2026.02 | 21.47 | 7.39 | 6.69 | 1.72 | 1.33 | |
| Audio2PhotoRealScheduler=DDIM, Denoising steps=50, CFG=22026.02 | 19.35 | — | 8.77 | 1.84 | 1.05 | |
| DyaDiTConditioning=CrossAttn2026.02 | 18.87 | 7.1 | 7.82 | 1.91 | 1.13 | |
| DyaDiTConditioning=w/o MD2026.02 | 18.34 | 7.37 | 6.88 | 1.75 | 1.29 | |
| ConvoFusionScheduler=DDIM, Denoising steps=50, CFG=22026.02 | 18.33 | — | 9.22 | 1.74 | 1.1 |