rem105-210 / difffashion Goto Github PK

I can't run your code, I used the versions of the torch 1.9.1 and torchvision 0.10.1 libraries because the libraries you specified are incompatible with each other and I have no idea how you started this project..

We will extend the method to more categories design (coming soon).

Combining the above two flowcharts, my understanding is that the method is only tuned in the inference phase?

That is, during denoising at the inference stage, only DINO-VIT (VIT_LOSS) is used to compute structure loss and appearance loss?

So the OceanBag Dataset is actually only used for evaluation, not for training?

How should I fine-tune it on other datasets if I expect the method to be more suitable for other datasets?

Looking forward to your reply, thank you so much~

I'm going to use this repo for pre-train, how much GPU memory will it cost and how long will it take to infer?

Thank you for your contribution! I have some questions about the code details and look forward to your answers! 1. Is the DDPM encoding mentioned in the article actually the process of adding noise to the image? 2. From line 825 to line 868 in gaussian_diffusion.py, the style reference image seems to have undergone 10 single-step denoising processes after noise addition, and each denoising process is reversed using the DDPM noise addition formula. What is the significance of this?
We look forward to your reply. Thank you again!

Hi!
Thank you for great work.