使用 OpenCV 和 Python 在 GPU 上 ZAEA23489CE3AA9B6406EBB28Z 的可行性

Question

I have to read images coming from a camera, add some shapes with OpenCV and return the image to a webpage with Flask. This application is running in Windows 10. The detailed steps are:

• Reading the images from a camera is done. I have a Python wrapper for the C++ library which seems to work well.
• Adding shapes with OpenCV is done. I use the OpenCV wrapper for Python.
• Returning the image with Flask is the problem. For that, I need to convert the image from raw to JPG or PNG format with cv2.imencode (the OpenCV Python wrapper). Nevertheless, this slows down the frame rate of the video to be shown on the website.

I've seen at some places, that OpenCV can be used to perform some operations over the GPU (Nvidia GeForce RTX). I've tried with cv2.UMat just before calling cv2.imencode but there is no difference.

And here is my question:

• should I try with more Python wrappers (as described here )?
• should I start learning C++? Will I need to migrate all the existing application from Python to C++ or just a wrapper? Can this be done in a few lines?

Answer 1

I finally compiled OpenCV-python with the GPU flags, with the needed CUDA flags. It took me weeks to install all the needed dependencies (image libraries, video libraries, Fortran, LINPAK, BLAS...).

The result is that the call to cv2.imencode is way faster and I can see the GPU consumption increased in the resources manager of Windows.