[Feature Request] - GPU implementation of HDR algorithms

[shatalinra]

Currently photo module have plenty of tonemapping and HDR algorithms but there is no GPU implementation. These algorithms are highly parallelizable and CUDA implementation can be much more faster than CPU implementation. The demand for this feature is demonstrated by these questions:

• https://stackoverflow.com/questions/61752270/how-to-generate-hdr-images-using-the-graphics-card-in-python
• https://answers.opencv.org/question/45387/hdr-imaging-gpu/
• https://answers.opencv.org/question/222780/real-time-hdr/
• https://answers.opencv.org/question/222761/how-to-speed-opencv-createmergemertens-function/

[alalek]

Try to utilize GPU through OpenCL (using UMat) - check tests code for details.

[iclowery]

I have a machine vision project which is currently taking around 5 seconds per HDR image creation using standard opencv - a cuda version would drastically reduce this and make the project feasible. Has this been addressed recently?