[Darknet] CPU usage optimization ./uselib API

Discussion started here with @rantgithub : https://github.com/opendatacam/opendatacam/issues/184#issuecomment-608226738

There is this issue on the darknet repo: https://github.com/AlexeyAB/darknet/issues/5170

Seems we can change a line on the darknet implementation:

in darknet/src/yolo_console_dll.cpp

while(!a_ptr.load()) std::this_thread::sleep_for(std::chrono::milliseconds(3));

to

while(!a_ptr.load()) std::this_thread::sleep_for(std::chrono::milliseconds(30));

@rantgithub , could you explain a bit more what would be the drawback of doing this ? I don’t really understand what is it doing ? Would it slow down the process for less high end devices ? How is this related to FPS ?

[rantgithub]

Hi

1. I am not sure why this is happening, but is a “big” problem, specially with more than 1 instance. Only solution i have found so far, is to buffer the stream using a 3rd party, because with the direct url m3u8, is just to fragmented. I will ask Alexyeb and see if he has any input

2. It is because i have touched many parts of the code on v2, in this way looking is how to put all together. I am testing this frames issue with v3 as well. I think we need a kind of internal buffer

[tdurand]

ok, so closing this