Hi,
I am trying to use the feature matches generated by LoFTR to help Nvidia Optical Flow’s. I read the Nvidia docs and the OpenCV cuda implementation, and I believe I have to use an array of the same dimensions as the input/reference frames. However, my external hints are not as dense, so my question is, does the hint buffer need to have a hint on every uv coordinate? Or I can leave those coordinates without a LoFTR match at zero?
Thanks!
Daniel