I know this is complicated and research is still going on but it also seems that there are many proprietary solutions for this. Is there any open source framework or library I could use to get this or at least the counting part of this done? I have tried OpenCV but the problem is that I am trying to implement this on CPU alone and the performance is really bad. Also, I am only able to detect human shapes up close but can't really find a way to detect small human shapes from pictures taken from say a camera two floors high. That is what guided me to the whole Tensorflow side. It'd be really great if someone could point me to the right direction on this. Cheers