about diffusion feature

[Lesleyblue]

Hello, I would like to ask what features are extracted from the trained UNet (1，256, 26, 64, 64). How are the 256 and 26 dimensions converted to 6656? Is it through the reshape method？

[chenning0115]

Thanks for concern about our work. You can find out from "https://github.com/chenning0115/spectraldiff_diffusion/blob/master/codes/feature_extract_unet.py" which is the code of feature extraction.