denoising-diffusion-pytorch
denoising-diffusion-pytorch copied to clipboard
U-Net architecture: Upsampling path
Hi Rosinality. I was going through the U-Net architecture you implemented for DDPM. I couldn't understand why you have added one extra resnet block in the upsampling path while not including it in the downsampling path (located in the 'model.py' file, specifically at line 346). Could you please clarify the reasoning behind this design choice?
up_layers = []
for i in reversed(range(n_block)):
for _ in range(n_res_blocks + 1):
channel_mult = channel * channel_multiplier[i]