search

dome272 / MaskGIT-pytorch

Pytorch implementation of MaskGIT: Masked Generative Image Transformer (https://arxiv.org/pdf/2202.04200.pdf)
MIT License
398 stars 34 forks source link

Pretrained model for VQGAN #11

Closed choyingw closed 2 years ago

choyingw commented 2 years ago

Thank you for the implementation! I would like to tune the second-stage transformer, but my VQGAN trained on Flickr landscape dataset is not so good. I see there is a load function of 'vq_flickr.pt' and you have much better landscape results. Could you kindly share that? Thanks!

out_2

dome272 commented 2 years ago

I dont have a pretrained flickr lying around anymore, but you may wanna take a look at concurrent work for the paper Make-A-Scene where we have released some ckpt. In the colab I also use rudalle and taming-transformer. All of these are general purpose models and should work pretty good for flickr. https://colab.research.google.com/drive/1SPyQ-epTsAOAu8BEohUokN4-b5RM_TnE?usp=sharing

choyingw commented 2 years ago

Thank you for the quick response!