I have extracted the optical flow along x and y axes. I want to pass them into a ConvNet. The thing I cannot understand is whether these should be two different input channels or should I combine them in some way, like stacking them, adding them or averaging them

Paper- Two-Stream Convolutional Networks for Action Recognition in Videos


Stack them so that it is an input with two channels. This is the standard approach.

  • $\begingroup$ Do you mean the optical flow images will have 6 channels, 3 for each ? $\endgroup$ – userlatebloomer Apr 17 '18 at 8:25
  • $\begingroup$ No, one channel is for the $x$ component of the flow, one channel is for the $y$ component of the flow. There are two channels in total. $\endgroup$ – shimao Apr 17 '18 at 8:58

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.