Reconstruction on MMBench 10 task sets 2 (unseen)
38.04PSNR (dB)Ours Post-FT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Ours Post-FTParams=102 M, Latent/frame=40962026.06 | 38.04 | 0.01 | 1.62 | |
| Wan 2.1 VAEParams=127 M, Latent/frame=40962026.06 | 36.62 | 0.01 | 0.17 | |
| Cosmos-CV8x8x8 (1.0)Params=106 M, Latent/frame=20482026.06 | 32.72 | 0.042 | 0.08 | |
| SD-VAE-MSEParams=84 M, Latent/frame=31362026.06 | 32.39 | 0.03 | 0.93 | |
| DC-AE-f32c32Params=323 M, Latent/frame=20482026.06 | 32.15 | 0.031 | 0.66 | |
| Ours BaseParams=102 M, Latent/frame=40962026.06 | 17.34 | 0.389 | 20.95 | |
| Ours Coverage-awareParams=102 M, Latent/frame=40962026.06 | 17.12 | 0.348 | 21.81 |