How to visualize different layers in FCN?

Is there a way to visualize activation maps for different layers in a Fully Convolutional Network? I assume since the up-sampling path outputs images anyways, this would be pretty easy.

To visualize outputs of a layer you need to create a function which takes in the input image and outputs that layer’s output. Then you just need to plot that function’s output. I have done visualization on the MNIST dataset, but you can use it on an FCN also.

Code link:-

@yashkatariya Do you know what happens if the image is RGB? Do you still need to visualize as gray image or can you use color. Any reference work you are aware of?

@s.s.o You can use color for visualization also. Using gray images gives a clear distinction of what the activation looks like. I have also done visualization on the CIFAR-10 dataset. While plotting the visualizations, for getting gray images, the parameter cmap is set to ‘gray’. You can try different strings or you can remove that to get color visualizations.

Code Link of CIFAR-10:-