1 | initial version |
Hi, unless I am mistaken it looks like the changes required to use cuDNN have not yet been merged into the main branch so you would need to build from the fork, more information can be found here.
Regarding python, I am not sure if the cuDNN options to access the DNN module are fully exposed to the python bindings in the fork but python bindings for the standard CUDA functions have been available since Aug 2018. Examples of most of the API calls are in test_cuda.py, for a full example see Accelerating OpenCV with CUDA streams in python.