pytorch-3dunet

pytorch-3dunet copied to clipboard

This PR implements a multicalss/mzltichannel aware `Normalize` transform functionality where `channelwise: true` can be specified so that the Normalize transform is applied to each channel seperately. In addition, `min_value` and...

jens-maus

Hi @wolny I came across this issue where the predictions are overwritten if multiple thresholds are used. A simple intermediate renaming should fix.

Josh-Talks

JulianYG

fix: remove redundant final_activation layer in model.py

1

Remove redundant final_activation layer in the forward() method of AbstractUNet. The layer is redundant because it is only applied during validation, however the loss function of segmentation models already applies...

JCasaraconn