search

songweige / TATS

Official PyTorch implementation of TATS: A Long Video Generation Framework with Time-Agnostic VQGAN and Time-Sensitive Transformer (ECCV 2022)
MIT License
263 stars 17 forks source link

Mugen checkpoint #15

Closed xiaoqian-shen closed 1 year ago

xiaoqian-shen commented 1 year ago

Great work and thanks for kindly releasing both codes and pretrained models! I could access all the checkpoints except the one on Mugen dataset. Could you please upload it to google drive instead of dropbox?

songweige commented 1 year ago

Thank you for your interests. Here are the gdrive links to the checkpoints: https://drive.google.com/drive/folders/1uJ9cY2XhwBBC1vEYcg8G3BV_G-OJ5Ugn?usp=sharing. Let me know if you still cannot access them!

xiaoqian-shen commented 1 year ago

Thanks very much! This issue can be closed.

xiaoqian-shen commented 1 year ago

Hi, is there an official FVD score for MUGEN dataset since I haven't seen it in the paper. I have generated videos with provided pretrained model but the fvd score is quite high. Probably because fvd is sensitive to preprocess or fps. Could you provide an official fvd score for fair comparison? Thanks very much!

songweige commented 1 year ago

Due to a version issue of the used MUGEN dataset mentioned here, we don't have the FVD values calculated.

I recommend checking the official codebase and table 3 in the paper for the baseline FVDs.