faced

Why we included this project

Faced is a small Python library that runs face detection on a plain laptop CPU at roughly five frames per second, which is enough for prototyping without a GPU. It chains two TensorFlow networks: a YOLO-style fully convolutional stage that proposes candidate boxes, followed by a second CNN that refines them. The author frames it as a proof of concept that a lightweight, single-purpose detector can beat heavyweight multiclass models on CPU, and the README includes a direct comparison with OpenCV's Haar cascades showing more robust results on angled and occluded faces. You get both a simple Python API and a command-line tool for images, video files, and live webcam feeds, and the code is easy to read if you want to see how a compact two-stage detector is built and trained on WIDER FACE. The caveat is honest: the author says it is not production-ready and the models are still being improved.