About the ddim_steps function in deffusion.py

Question

About the ddim_steps function in deffusion.py

noway2beatme opened this issue 2 years ago · 2 comments

x_cond[0][0] = [torch.stack([i * feat[0] + (1 - i) * feat[-1] for i in torch.linspace(0, 1, len(feat))], 0) for feat in x_cond[0][0]]

What does this line of computation do?

Answer 1 · 2023-04-18T12:46:10.000Z

Thank you!

Answer 2 · 2023-04-19T13:23:04.000Z

I appreciate you bringing this to my attention. This line was supposed to be commented out, but I must have missed it and it went unnoticed. However, this line had no impact on the inference or in the colab demo file where bs is set to 1.

The DDPM sampling is okay.