I'm a bit confused about this. Assume I have a CNN network with two branches:

  1. Top
  2. Bottom

The top branch outputs a feature vector of shape 1x1x1x10 (batch, h, w, c) The bottom branch outputs a feature vector of shape (1, 10, 10, 10).

I want to use the top feature vector as a convolutional filter, and convolve it with the bottom feature vector. I can do this in pytorch with the "functional.Conv2D" function, the problem is, I don't know how back-prop works in this case (will it be unstable?) since the output feature is a now a parameter as well, do I need to stop gradients or do something else in this case to backprop correctly?


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.