Specify CUDA stream for DNN evaluation

asked 2020-08-03 10:56:29 -0600

rgov gravatar image

I have a CUDA-accelerated pipeline for processing an image. At the end of the pipeline I have to download the processed image from the GPU, then use the cv.dnn.blobFromImage API and send it back to the CUDA DNN backend.

Can I connect these so that the image never leaves the GPU? Can I specify that it use the same stream?

edit retag flag offensive close merge delete