BlazeFace in Python
BlazeFace is a fast, light-weight face detector from Google Research. Read more, Paper on arXiv
A pretrained model is available as part of Google's MediaPipe framework.
Besides a bounding box, BlazeFace also predicts 6 keypoints for face landmarks (2x eyes, 2x ears, nose, mouth).
Because BlazeFace is designed for use on mobile devices, the pretrained model is in TFLite format. However, I wanted to use it from PyTorch and so I converted it.
NOTE: The MediaPipe model is slightly different from the model described in the BlazeFace paper. It uses depthwise convolutions with a 3x3 kernel, not 5x5. And it only uses "single" BlazeBlocks, not "double" ones.
The BlazePaper paper mentions that there are two versions of the model, one for the front-facing camera and one for the back-facing camera. This repo includes only the