List of important approaches:

1. Viola-Jones Detector (CVPR, 2001)

   • Available in OpenCV
   • A tutorial can be found here: https://docs.opencv.org/3.4/db/d28/tutorial_cascade_classifier.html

2. MTCNN (2016)

   • Original code: https://github.com/kpzhang93/MTCNN_face_detection_alignment
   • The model was originally developed in Caffe
   • Repository with OpenCV and C++: https://github.com/ksachdeva/opencv-mtcnn
   • Repository with OpenCV and Python: https://github.com/linxiaohui/mtcnn-opencv

3. SSD (ECCV, 2016)

   • The official code repository is here: https://github.com/weiliu89/caffe/tree/ssd
   • OpenCV integrates a pretrained model that can be found here: https://github.com/opencv/opencv/tree/master/samples/dnn/face_detector
   • Instructions how the SSD model was trained are also in the OpenCV repository
   • The weights probably stem from this repository: https://github.com/sr6033/face-detection-with-OpenCV-and-DNN

4. RetinaFace (CVPR, 2020)

   • RetinaFace is originally trained in MXNet and the code can be found here: https://github.com/deepinsight/insightface/tree/master/detection/retinaface
   • For usage in inference, I would use this repository: https://github.com/Charrin/RetinaFace-Cpp
   • The Caffe model can probably be loaded with OpenCV

5. YOLO5-Face (ArXiV, 2022)

   • The official code repository is here: https://github.com/deepcam-cn/yolov5-face
   • The model is trained on WiderFace dataset with Pytorch
   • The YoloV8-Face repository is here: https://github.com/derronqi/yolov8-face/tree/main
   • Converted ONNX models for yolov8 can be found here: https://github.com/hpc203/yolov8-face-landmarks-opencv-dnn

1. Install miniconda

2. Create conda environment with

   conda env create -n fdetect -f environment.yml

3. Install pip requirements

   conda activate fdetect
   pip install -r requirements.txt

4. Install pre-commit

   pre-commit install

1. Face detection:

   python detect_faces.py (-h)

2. Face alignment:

   python detect_align.py (-h)

Use -h flag to show options

• Fix performance discrepancy between src/mtcnn_onnx.py and src/mtcnn_caffe.py