I am trying to accelerate some image processing that requires tasks not currently available in any one python CUDA module. For example, [`cupy`](https://cupy.chainer.org/) has a lot of `numpy`/`scipy` functions that are not available in `OpenCV`. I'm curious if anyone has had success passing the data on the GPU between different code bases, instead of spending a lot of compute time passing back and forth between CPU and GPU to do the conversions? I'm specifically interested in `cupy` but could use any experiences with other codes as a start.
↧