I would like to know if it is possible to read a video directly from the GPU with Open CV, without capture the image with CPU and then upload it to the GPU.
Could you suggest some tutorials to follow?
We don’t know much about how openCV implements GPU Mat, but your usecase seems require memory copy.
Could you share your target? What do you want to do with opencv?
I’m interested on glass-to-glass latency calculation about onboard camera and the ways to reduce latency. I would like to use the Jetson’s GPU and not CPU through OpenCV.
Please try our mmapi sample. Those samples are basically zero copy on GPU with camera use.