Face-api.js: Investigating in YOLO (v2 or v3) for face detection

[justadudewhohacks]

It might be worth it to also investigate in yolo v2 or v3 as an alternative face detector. From the looks I would tend to yolo v2, as it is more leaned towards achieving high fps. Furthermore, there seem to be some already trained models for face detection with yolo v2 out there, which would make it easier to get started.

[justadudewhohacks]

Just merged an experimental version of tiny yolo v2 for face detection. I think this one could see a huge performance speedup by replacing the conv2ds with depthwise seperable convolutions, but therefore the model has to be retrained.

[justadudewhohacks]

Currently training the tiny yolo v2 model using depthwise separable convolutions from scratch. Speed up over regular convolutions is huge. The current model already works pretty well for face tracking and even runs on mobile chrome. First version to play around with can be tested here: https://yolosep.herokuapp.com/