Just a quick question, I noticed that Sore-CAM is implemented in a way that supports only ALexNet. In order to support other pretrained models like Resnet, should I change the forwad_convolutions function and extract the last conv layer before pooling or should I change anything else?
@dimimal Yes, as the network structures are different, you need to extract the feature map of ResNet by modifying several lines of code. Here are some references that may be helpful to you.
Hi,
Just a quick question, I noticed that Sore-CAM is implemented in a way that supports only ALexNet. In order to support other pretrained models like Resnet, should I change the forwad_convolutions function and extract the last conv layer before pooling or should I change anything else?