Hello everyone,
I’ve found a student who’s interested in taking part in a GSoC on improving DISFlow in OpenCV.
The first point I’d like to see addressed in this context is the generalisation of Charbonier’s loss function to the one developed in this article “A General and Adaptive Robust Loss Function”, in variational_refinement.
The second point would be to allow DISFlow, and therefore template matching and variational_refinement, to be able to work along an axis so that we can use these methods in the stereo context, because the methods used in OpenCV to obtain the disparity map are dated.
Do you know how to go about making this proposal?