Xiyue-Wang / RetCCL

GNU General Public License v3.0
93 stars 8 forks source link

GradCam on CCL-model #1

Open janniehues opened 2 years ago

janniehues commented 2 years ago

Hello there,

I am a medical student and am using your model to extract features from histological images. These features are then analysed with an FC-layer on top, to learn particular classes. Your model is working really great and I am very grateful that you made it public! To visualize what the trained model focuses on I would like to apply CAM/GradCam next. Now that you use a CCL-model I am not sure how to correctly extract the feature map relevant for CAM/GradCam in this case (in normal ResNets it is the layer before avgpool). Could you maybe help me with this? Usually I get the GradCam by calculating the forward pass to the relevant layer, call a retain_grad on that layer, and then forward the last layers.

Thank you very much, Jan

Xiyue-Wang commented 2 years ago

Thank you very much for your attention to our work. This is indeed used by many people. It is a huge lead over imgnet and is more suitable for pathological images. This article is being major revised. We are busy revising the article. If accepted, we will update these