This repository is synced

OpenCV-Yolov3

I will likely introduce some changes later. It's a pretty cool project to have fun with. I might want to add some kind of python prerequisites file, then try modifying the drawing algorithm.

It's nothing really fancy (although, it's pretty cool) and currently more of an implementation of an article/guide I managed to stumble upon.

Based on

• Nandini Bansal's medium story with GitHub Project
• My own modifications

Changelog TO-DO/IN-PROGRESS

• configuration parameters on top of the file (IN-PROGRESS)
• configuration of the backend's target (CPU, GPU) (Done, not tested)
• quit key on 'Q' (Done)
• changeable/adjustable font/rectangle colors (TO-DO)
• FPS counter (TO-DO)
• possible split of functions into separate files (TO-DO)
• logging of detected objects (FUTURE)

I introduced my own changes, basing on public domain resources (including Chinese websites).

I am actually not sure if I could manage to port it to use GPU instead. I believe that some frameworks offered CUDA support, but I am running on Radeon.

Performance:

Ryzen 2600X - 60-70% CPU use at (measuring by eye) 10-15FPS

Dependencies

• pip install opencv-python
• pip install numpy
• pip install argpars
• pip install time

Network dependencies

• CFG file at: https://github.com/pjreddie/darknet/blob/master/cfg/yolov3.cfg
• Coco labels at: https://github.com/pjreddie/darknet/blob/master/data/coco.names
• Weights at: https://pjreddie.com/media/files/yolov3.weights

Example