Conversation

Replying to
Hi Andrey, I am really impressed by your work and am trying hard to reproduce it. However I found that the loss gradient guidance toward the latents does not work as expected. Specifically, would not it be contradictory to "edge prediction based on noisy image is as com
Replying to
complex as the original DDPM denoising". In that case, wouldn't the gradients toward the first few channels of the input to your edge predictor be very trivial to do any meaning guidance?
1
Show additional replies, including those that may contain offensive content
Show