MTCNN

PyTorch implementation of inference stage of face detection algorithm described in
Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks.

Why this projects

mtcnn-pytorch This is the most popular pytorch implementation of mtcnn. There are some disadvantages I found when using it for real-time detection task.

• Mix torch operation and numpy operation together, which resulting in slow inference speed (Cannot run on GPU).
• Based on the old version of pytorch (0.2).

So I create this project and make some improvments:

• Transfer all numpy operation to torch operation, so that it can benefit from GPU acceleration.
• Automatic run on 'CPU' or 'GPU'.
• Based on the latest version of pytorch (1.3)
• Real-time face tracking

Installation

1. Create virtual environment

   # conda create -n face_detection
   

2. Activate virtual environment

   # source activate face_detection
   

3. Install PyTorch

   (face_detection)# conda install pytorch torchvision cudatoolkit=10.1 -c pytorch
   

4. Install Jupyter Notebook

   (face_detection)# conda install jupyter notebook
   

5. Install opencv

   (face_detection)# pip install opencv-python
   

How to use it

Just download the repository and then do this

import sys
import os
sys.path.append(os.pardir)
from importlib import import_module
import cv2
from src.detect import FaceDetector

# import camera driver
if os.environ.get('CAMERA'):
    Camera = import_module('camera_' + os.environ['CAMERA']).Camera
else:
    from camera import Camera

if __name__ == "__main__":
    detector = FaceDetector()
    
    while True:
        frame = Camera().get_frame()
        image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)        
        bounding_boxes = detector.detect(image)

        for i in range(len(bounding_boxes)):
            cv2.rectangle(frame, (int(bounding_boxes[i][0]), int(bounding_boxes[i][1])),
                         (int(bounding_boxes[i][2]), int(bounding_boxes[i][3])), (255, 0, 0), 2)

        cv2.imshow('capture', frame)
        key = cv2.waitKey(1)
        if key & 0xFF == ord('q'):
            break

or just run:

(face_detection)# cd MTCNN_face_detection_alignment_pytorch/src/
(face_detection)# CAMERA=opencv python test.py

Tutorial

Detect step by step

Credit

This implementation is heavily inspired by:

• TropComplique/mtcnn-pytorch
• faciallab/FaceDetector

Citation

@article{7553523,
    author={K. Zhang and Z. Zhang and Z. Li and Y. Qiao}, 
    journal={IEEE Signal Processing Letters}, 
    title={Joint Face Detection and Alignment Using Multitask Cascaded Convolutional Networks}, 
    year={2016}, 
    volume={23}, 
    number={10}, 
    pages={1499-1503}, 
    keywords={Benchmark testing;Computer architecture;Convolution;Detectors;Face;Face detection;Training;Cascaded convolutional neural network (CNN);face alignment;face detection}, 
    doi={10.1109/LSP.2016.2603342}, 
    ISSN={1070-9908}, 
    month={Oct}
}