Tks for posting, great sharing. I got some info that without a GPU, it may take 1-2s per frame, What If I only like to identify certain categories ? (say, only car, and motocycles), could it increase the speed (without GPU) ?
The inference will take the same time since the only difference is the number of categories that you trained the network with. The architecture of the network remains the same, thus the elapsed time from input to output will also remain the same.