FoundationVision / OmniTokenizer

[NeurIPS 2024]OmniTokenizer: one model and one weight for image-video joint tokenization.
https://www.wangjunke.info/OmniTokenizer/
MIT License
263 stars 7 forks source link

Question about the gan_feat_loss #9

Closed hyc9 closed 4 months ago

hyc9 commented 4 months ago

Thank you for open-sourcing such excellent work! In your code, I noticed that while training the autoencoder, the GAN loss includes both aeloss and gan_feat_loss. To my knowledge, typically only aeloss is used in the training of the generator. Therefore, I wonder whether there's reference for the gan_feat_loss you used. I greatly appreciate your help.

wdrink commented 4 months ago

Please refer to the TATS paper https://arxiv.org/abs/2204.03638