Dominic Rampas
Dominic Rampas
With this option you can predict x_0 directly from x_t. However, this was shown in the papers to not work well. See https://arxiv.org/pdf/1503.03585.pdf on page 2.
in sample_conditional.py :)
Hey, this is work in progress and we have not yet implemented it. We are currently training VQIMG and are slowly starting to implement everything for the transformer.
Hey there and thanks for the nice words, Unfortunately, I can not say much with just this error line. It could be many things. Can you give me more context...
Hey this is an awesome addition and thank you so much for the work. The only problem I have is that people might be confused who are coming from the...
Very very cool. Thank you!! I added a reference to your repo and blog in the readme
Hello, the first stage of the VQGAN implementation might not be perfect. The embedding dimension might be a worthy thing to try, I used 256 because my machine couldnt do...
> where it says all models have 24 layers, 8 attention heads, 768 embedding dimensions and 3072 hidden dimensions This is for the second stage of the VQGAN, so for...
No problem, also just drop your adjustments you made hear if you found something to work for your case. Maybe others can benefit from it too.
Unfortunately, I could not even train a model yet, because my compute ressources are allocated otherwise at the moment. Did you use your own codebase? If so, did you publish...