Conditional Video Generation on Kinetics-600 (test)
2.3FVD (50k)SiD2
Evaluation results
| Method | Links | |
|---|---|---|
| SiD2sigmoid bias=0, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 2.3 | |
| SiD2 (sigmoid bias = -2)sigmoid bias=-2, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 2.5 | |
| SiD2 (sigmoid bias = +2)sigmoid bias=2, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 3.2 | |
| W.A.L.Tcategory=latent diffusion, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 3.3 | |
| MAGVITv2category=latent diffusion, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 4.3 | |
| Rolling Diffusioncategory=pixel diffusion, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 5.2 | |
| SiD2 (eps-mse)loss=eps-mse, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 5.7 | |
| MAGVITcategory=latent diffusion, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 9.9 | |
| RINcategory=pixel diffusion, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 10.8 | |
| Video Diffusion Modelscategory=pixel diffusion, Resolution=128x128 (resized to 64x64), Conditioning frames=5, Generated frames=112024.10 | 16.6 |