It seems that extrating the dlib-19.22.0-py3.6-linux-aarch64.egg to the ~/.local/lib/python3.6/site-packages/ and putting ./dlib-19.22.0-py3.6-linux-aarch64.egg into the easy-install.pth, located in ~/.local/lib/python3.6/site-packages/, enables me to have my python environment recognizing the dlib-19.22.0 version.
If in this file I swap ./dlib-19.22.0-py3.6-linux-aarch64.egg to ./dlib-19.21.0-py3.6-linux-aarch64.egg it makes me use the dlib-19.21.0 version and by swapping back by putting ./dlib-19.22.0-py3.6-linux-aarch64.egg in the file, it leads to the use of the dlib-19.22.0 version.
However, by using the new version, dlib-19.22.0, GPU version, I do not find if faster than the previous one, no matter how many trackers are present in the frame. Is it normal?
When I check with the jtop command in the terminal, I found that the GPU version has almost the same result than the CPU version, like in the picture above, with the GPU going from 0 to 10% sometimes, but usually near 0%.
Are all I did right? Is there some things to change when I create DLIB trackers, like explicitly saying dlib.cuda.correlation_tracker() instead of dlib.correlation_tracker() ?
Regards.