About the data preprocess
Ashliquid opened this issue · 1 comments
Ashliquid commented
Is your algorithm sensitive to data preprocessing methods? I did the min-max normalization and the results were not ideal.
I retrained the diffusion model in the algorithm on your weights file.
Aboriginer commented
Hi @Ashliquid, sorry for the late reply.
If you change the normalization, I think you should tune hyperparameters (sampling.snr, sampling.mse, and sampling.corrector_mse) during the inference time to achieve better results.