When I upload two pictures(both are 1920x1080) from Mat to GpuMat, I run twice and it takes an average of 50 ms per picture in the second time! Could someone tell me this is normal or what’s wrong. If somebody helps me I will be very gratefull.
Device: Jetson Xavier NX
OpenCV version: 4.5.2
CUDA version : 10.2
There’s the code for example :
I think the first thing you would want to do is determine the “speed of light”-performance. Presumably assigning a Mat to a GpuMat involves copying the data? Combine the relevant HW bandwidth with the size of each picture to compute how fast it could be copied in the theoretical case scenario.