What about a specific subset of features from a layer?

[taimoor6864]

Hey, I hope you are well.
I am trying to run your reconstruction for the an intermediate layer of the VGG-16 (ReLU2_2) which has 128 channels. Is there any possibility to use a subset of those 128 channels (like say 40 of my choice, based on a criteria that I develop) to reconstruct the input image. I see in your paper that you did divide the channels into two parts in one experiment. If possible, some directions on how to achieve this would be greatly appreciated. Thank you for your time.