I'm currently experimenting with Resnets and I'd like to visualize the convolutional layers for that. I came across this Keras Blog to perform an activation maximization of the convolutional layers:


The code used for the blog post is based on the pretrained VGG model from Keras. I tried to adapt the code so that it loads the pretrained Resnet50 from Keras. Apart from that I left the code as it comes.

I primarly visualized the beginning convolutional layers and the last convolutional layers of the Resnet50. However, they don't look the same as in the VGG model presented in the blog above, rather a bit more noisy, especially at the last layers.

I put some visualizations of the Resnet50 below (from the very first layers to the last ones):

Are these reasonable visualizations in the case of the Resnet50 or did I miss something that I need to do, when switching from the VGG model to the Resnet50 model but using the same code that is provided by Keras?

enter image description here

enter image description here

enter image description here


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.