Apply MTCNN through keras with pre-trained weights
1 Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks .pdf
This is the lecture of multi-task cascaded convolution networks(MTCNN) which declares the architecture of the networks, the training process and the comparison with other networks.
2 image dircetory
This dircetory contains the test images and the output images , the later are in out directory.
3 model weights directory
This directory contains the MTCNN weights including 12net.h5(pnet.h5), 24net.h5(rnet.h5) and 48net.h5(onet.h5)
4 MTCNN.py
The architecture of the MTCNN, containing 3 function create_Pnet, create_Rnet and create_Onet
5 mtcnn_utils.py
Some assistant function to realize the image pyramid change, post process after each network, non-max-suppression and image shape change, et al. There are 2 functions to do NMS and image shape change, and both have similiar performance.
6 face_detection.py Put the upper functions together to do the face detection.
7 Detector.py (Detector.ipynb)
The detection function, you can pass a image to function face_detection() and then use openCV to display the final output and store it into image/out.
python == 3.7.4
numpy == 1.18.1
opencv-python == 4.2.0.32
tensorflow == 1.15.0
Keras == 2.1.0