Video Reconstruction on High-motion video 160 videos (val)
19.5rFVDWan 2.1 (patchify)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Wan 2.1 (patchify)Tokenizer Type=Continuous, Latent Channel=64, Spatial Stride=16, Temporal Stride=4, Channel Bits=16, Compress Ratio=242026.04 | 19.5 | 0.058 | 92.9 | 34.1 | |
| Wan 2.2 (patchify)Tokenizer Type=Continuous, Latent Channel=48, Spatial Stride=16, Temporal Stride=4, Channel Bits=16, Compress Ratio=272026.04 | 22.6 | 0.052 | 93.2 | 34.54 | |
| HBQ (M=4, tune lambda_GAN)Tokenizer Type=Discrete, Latent Channel=64, Spatial Stride=16, Temporal Stride=4, Channel Bits=4, Compress Ratio=962026.04 | 30.1 | 0.078 | 92.8 | 33.98 | |
| HBQ (w/o quant)Tokenizer Type=Continuous, Latent Channel=64, Spatial Stride=16, Temporal Stride=4, Channel Bits=16, Compress Ratio=242026.04 | 43.2 | 0.078 | 93.5 | 34.79 | |
| HBQ (M=4)Tokenizer Type=Discrete, Latent Channel=64, Spatial Stride=16, Temporal Stride=4, Channel Bits=4, Compress Ratio=962026.04 | 50.6 | 0.084 | 93 | 33.97 | |
| HBQ (w/o quant)Tokenizer Type=Continuous, Latent Channel=16, Spatial Stride=16, Temporal Stride=4, Channel Bits=16, Compress Ratio=962026.04 | 144.6 | 0.141 | 87.9 | 31.14 | |
| HBQ (M=8)Tokenizer Type=Discrete, Latent Channel=16, Spatial Stride=16, Temporal Stride=4, Channel Bits=8, Compress Ratio=1922026.04 | 144.9 | 0.141 | 87.9 | 31.1 | |
| HBQ (M=6)Tokenizer Type=Discrete, Latent Channel=16, Spatial Stride=16, Temporal Stride=4, Channel Bits=6, Compress Ratio=2562026.04 | 148.8 | 0.142 | 87.8 | 30.98 | |
| HBQ (M=4)Tokenizer Type=Discrete, Latent Channel=16, Spatial Stride=16, Temporal Stride=4, Channel Bits=4, Compress Ratio=3842026.04 | 163.6 | 0.148 | 87.2 | 30.4 |