I'm assuming that if I build with CUDA support to use the GPU that I should see performance increase through the python modules as well as these are essentially just calling the C/C++ functions correct?
No. CUDA optimization is explicit. OpenCV has separate module opencv_gpu, that contains different algorithms implemented via CUDA. User should explicitly call functions from opencv_gpu module to run code on GPU, for example:
cv::Mat img = cv::imread(...);
cv::gpu::GpuMat d_img;
d_img.upload(img); // copy data to GPU memory
cv::gpu::GpuMat d_edges;
cv::gpu::Canny(d_img, d_edges, 50, 100); // call GPU implementation
cv::Mat edges;
d_edges.download(edges); // copy data from GPU to CPU memory
And, unfortunately, there is no Python wrappers for opencv_gpu module.
Asked: 2013-10-14 09:28:12 -0600
Seen: 835 times
Last updated: Oct 14 '13
