Face detection with OpenCV and deep learning - PyImageSearch

Abstraction: OpenCV hidden deep learning SSD-ResNet face detector usage

Key points:

• OpenCV ships a "hidden" DNN-based face detector since version 3.3, using SSD framework with ResNet base network (not MobileNet)
• Model requires two Caffe files: a .prototxt architecture file and a .caffemodel weights file (res10_300x300_ssd_iter_140000.caffemodel)
• Input images are preprocessed into 300x300 blobs with mean subtraction (104.0, 177.0, 123.0) via cv2.dnn.blobFromImage
• Outperforms Haar cascades significantly: detects faces at angles and in challenging lighting (e.g., 86.81% confidence in dark venue)
• Works for both static images and live video streams; real-time capable unlike dlib's CNN detector without GPU
• Alternative detectors compared: Haar cascades (fastest but least accurate), dlib HOG/CNN, and MTCNN

Connections: Opencv · Face Detection · Object Detection

Source: https://www.pyimagesearch.com/2018/02/26/face-detection-with-opencv-and-deep-learning/