This is a tiny yolo face detector trained on FDDB+Dlib dataset. It was trained on a GTX1080 for about 82k iterations. It runs fast at 112 fps on GTX1080 which is more than enough for realtime usage.
Clone the repo
>git clone https://github.com/azmathmoosa/azFace.gitCD into the repo
>cd azFace
Launch yolo_console_dll.exe followed by path to video/image
>yolo_console_dll.exe C:\random\video.mp4
Or use the darknet executable
>darknet.exe detector demo net_cfg\azface.data net_cfg\tiny-yolo-azface-fddb.cfg weights\tiny-yolo-azface-fddb_82000.weights C:\Dataset\random\crowd.mp4
- Clone this repo and the darknet repo.
- Follow the instructions of darknet to build it
- After building use the provided cfg and weight files like so
./darknet detector demo net_cfg/azface.data net_cfg/tiny-yolo-azface-fddb.cfg weights/tiny-yolo-azface-fddb_82000.weights /path/to/my/video.mp4
To record a video use this command
>darknet.exe detector demo net_cfg\azface.data net_cfg\tiny-yolo-azface-fddb.cfg weights\tiny-yolo-azface-fddb_82000.weights C:\Dataset\random\crowd.mp4 -out_filename test.avi
You will need DivX codec installed to record and VLC to play the video.
This work is licensed under LGPLv3. Please attribute to the author incase you find this work useful.
In my spare time I offer consultation services for deep learning projects. If you need assistance for your projects feel free to reach me at a z m a t h m o o s a @ g m a i l dot c o m