Is it possible to use Numba for opencv to speed up say disparity computation?
opencv is a c++ library (with python wrappers)
so, unfortunately, you can only try to improve the python parts of your program
(e.g. if you have code like here)
I didn’t try Numba with OpenCV, but OpenCV images in Python are in fact Numpy arrays. So you can use those arrays in Numba.
That said, OpenCV has hardware accelerated stereo disparity computation: OpenCV: Stereo Correspondence, you can use that directly without needing to reimplement everything in Numba.
[small note: the python bindings of the OpenCV Cuda module are not documented, so you’ll need to fillde around to get your parameters right, but it works]