search

KidsWithTokens / MedSegDiff

Medical Image Segmentation with Diffusion Model
MIT License
980 stars 147 forks source link

The out_channels of the model #91

Closed YonghanLU closed 1 year ago

YonghanLU commented 1 year ago

Why the out_channels of the model is 2 ? I do not understand , the output for labels should be 1 channel if not learn sigma.

WuJunde commented 1 year ago

check this issue: https://github.com/WuJunde/MedSegDiff/issues/3

YonghanLU commented 1 year ago

Thanks , but in the 'def diffusion_defaults()' of your code , the learn_sigma is False , then the model_var_type will be gd.ModelVarType.FIXED_LARGE at 'def create_gaussian_diffusion' in script.util.

YonghanLU commented 1 year ago

Should I change learn_sigma to True ? When False , I found it also works , but I think this may be redundant,

YonghanLU commented 1 year ago

Sorry , I review your README , I found I ignored "learn_sigma True" .You are right , thanks a lot again.