Face-Detector with Landmark

This project showcases a lightweight face detector with keypoint detection, designed for mobile deployment. The model is based on Retinaface-mobile0.25, converted into ncnn python, opencv onnx, and pytorch python, providing a slim, efficient face detector suitable for edge computing.

Key Features:

• Face-Detector-1MB slim with 5 key points for face detection.
• Retinaface-mobile0.25 model converted into multiple formats: ncnn python, opencv onnx, pytorch.
• Support for onnx export.
• Network parameter and FLOP calculation for performance optimization.
• Modified anchor size of Retinaface-mobile0.25 to enhance suitability for mobile and edge devices.
• Reimplementation of Face-Detector-1MB, adding key point detection and ncnn C++ deployment.
• Enhanced accuracy compared to original models, making it ideal for real-time applications.

Technologies and Dependencies:

• Ubuntu 18.04
• Python 3.7
• OpenCV
• NumPy

Accuracy:

Widerface Test (Single-scale Input Resolution: 320x240)

Method	Easy	Medium	Hard
libfacedetection v1 (caffe)	0.65	0.5	0.233
libfacedetection v2 (caffe)	0.714	0.585	0.306
version-slim (original)	0.765	0.662	0.385
version-RFB (original)	0.784	0.688	0.418
version-slim (our)	0.795	0.683	0.345
version-RFB (our)	0.814	0.710	0.363
Retinaface-Mobilenet-0.25 (our)	0.811	0.697	0.376

Widerface Test (Single-scale Input Resolution: 640x480)

Method	Easy	Medium	Hard
libfacedetection v1 (caffe)	0.741	0.683	0.421
libfacedetection v2 (caffe)	0.773	0.718	0.485
version-slim (original)	0.757	0.721	0.511
version-RFB (original)	0.851	0.81	0.541
version-slim (our)	0.850	0.808	0.595
version-RFB (our)	0.865	0.828	0.622
Retinaface-Mobilenet-0.25 (our)	0.873	0.836	0.638

Note: When testing, the long side is either 320 or 640, and the image is scaled in equal proportions.

References:

• FaceBoxes
• Retinaface (mxnet)
• Retinaface (pytorch)
• Ultra-Light-Fast-Generic-Face-Detector-1MB

You can view the full project and source code on GitHub: Face-Detector with Landmark