Inquiry About Comparison with WF-VAE

[LAW1223]

Hello, I’ve been following your work, and I find it particularly interesting and helpful. May I ask two questions？

1. I'd like to learn more about how your work compares to the approach used in WF-VAE (https://github.com/PKU-YuanGroup/WF-VAE). Would you be willing to share some insights on this?

2. How about the inference speed?

[annitang1997]

Thank you for your interest in our work.

1. We have tested the open-source model of WF-VAE, and the comparison results under the same setting (causal model, video compression ratio: 4×8×8, input shape: 17×256×256, testing data: MCL-JCV, 30 FPS) is as follows:

Method	Param.	PSNR	SSIM	LPIPS	FVD
WF-VAE-L-16chn	317M	33.76	0.928	0.091	90.8
VidTok-16chn	157M	35.04	0.942	0.047	78.9

2. As for the inference time, we will provide updates on it later.