Question on the normalization method for calculating class activation maps.

[Minxiangliu]

Thank you very much for providing a great graphical explanation tool, MONAI. I have a question regarding it. I'm currently using the monai.visualize.CAM tool to explain my model, which is torchvision.models.efficientnet_b2. Everything is running smoothly, but while understanding the code, I noticed in monai.visualize.class_activation_maps.default_normalizer it mentions This will flip magnitudes. Could someone explain why? Why is this necessary?
Thanks in advance!

[KumoLiu]

Hi @Minxiangliu,
It's for the visualization, no special reason I think.
https://github.com/Project-MONAI/tutorials/blob/main/modules/interpretability/class_lung_lesion.ipynb
Thanks.