mehdidc / feed_forward_vqgan_clip

Feed forward VQGAN-CLIP model, where the goal is to eliminate the need for optimizing the latent space of VQGAN for each input prompt
MIT License
136 stars 18 forks source link

Unavailable and broken links #15

Closed woctezuma closed 2 years ago

woctezuma commented 2 years ago

When I run the notebook, some links seem unavailable. I don't know why this happens, because it seems that I can manually download the files in my web browser.

Unavailable links

Moreover, the links in the README are broken.

Broken links

mehdidc commented 2 years ago

Hi @woctezuma, thanks for raising this issue, could you please try again the notebook, it should work now. I will also change the links in the README.

EDIT: it is weird though that you could manually download them fine in your browser, so this could be another problem. I tried to use the notebook by logging using another gmail account (rather than the one the files belong to), and it did work, not sure what is happening.

afiaka87 commented 2 years ago

@mehdidc This happened with the original CUB dataset for the dalle-pytorch colab. It was on lucidrains gdrive. I think Google makes them more-restrictive if it gets flagged or something.

mehdidc commented 2 years ago

Ah I see, thanks, so perhaps it's better to switch to e.g dropbox

woctezuma commented 2 years ago

I don't know if it could help, but I have seen some projects share model checkpoints in the "release" section of the Github project.

For instance: https://github.com/neuralchen/SimSwap/releases which can be accessed by clicking here: Access to releases

mehdidc commented 2 years ago

Ah cool thanks @woctezuma , looks like a good solution (there is no total size limit according to the github doc), in fact now I recall @afiaka87 used it for the blog captions model actually.

mehdidc commented 2 years ago

Ok I updated the links using a github release, please try again, thanks for the tip. It's nice, the only issue is that it does not work for files > 2GB.

woctezuma commented 2 years ago

It works! Thank you!