evaluation IoU, Dice, F1 score in per class;

[heikeyuhuajia]

Hello, I'm recently conducting comparison experiments. Some of these experiments are conducted through the MMsegmentation framework, which has evaluation metrics including

IoU, Dice, F1 scores for different categories,

and I would like to know these metrics for Meddiff, so I wrote this script for evaluation.

[heikeyuhuajia]

I also have a question: can I use the binary (0, 1) pred label predicted by other segmentation models and compare the experimental results by

segmentation_env.py

file? It should be noted that the traditional model does not pass the ensemble technique.

[WuJunde]

Awesome! Thanks for your contribution! b.t.w, I think segmentation_env.py works for the binary masks ( the thresholding becomes meaningless but will not affect the final result.)