人脸口罩检测（中文）/ Face mask detection (English)

Update 2021/01/01: Faster R-CNN based Face Mask Detector

We recommend to use anaconda to create a python3 environment to manage the pytorch-GPU environment.
You can use the following commands to configure your environment:

conda create -n {your environment name} python=3.7

Then anaconda will solve the dependencies automatically for you. (Make sure you have successfully installed the NVIDIA driver.)

Then you need to install following dependencies in your conda environment:

pytorch-GPU or pytorch > 1.0  
python-opencv > 3.0  
torchvision
numpy

1. Download the AIZOO Face Mask Detection Dataset (or you can use your own dataset, make sure in same format) (link) and copy the dataset AIZOO into the root folder as /Face-Mask-Detection/AIZOO. The files in AIZOO are as follow:

   /AIZOO/
    ├── train
    ├── val
    └── readme.md

2.Process the Dataset by running 'dataset_copy.py'. In our demo, we use 800 images for training and 500 images for validation.
Then processed train folder and val folder will show in the root folder as follow:

   /ROOT/  
    ├── train  
    │   ├── Annotations  
    │   ├── JPEGImages  
    │   └── train.txt  
    └── val  
        ├── Annotations  
        ├── JPEGImages  
        └── val.txt  

note: you can choose how many images you want to use for training and validation by revising 'dataset_copy.py'.  

3. Traing the Faster R-CNN based Face Mask Detection Model. Run train_faster_rcnn.py (you can set customized parameters)

1. After the trained models are saved in checkpoints_faster_rcnn, you can run evaluation_faster_rcnn to calculate mAP for both Face With mask and Face Without Mask.
2. After the trained models are saved in checkpoints_faster_rcnn, you can run demo_faster_rcnn.py to visualize the detection result by changing demo.jpg.

epoches = 10, Adam Optimizer, learning rate = 0.0001
800 images (400: with mask 400: without mask) for training
500 images (250: with mask 250: without mask) for testing

Face Without Mask

Face With Mask

https://github.com/AIZOOTech/FaceMaskDetection
https://github.com/aky15/AIZOO_torch
Ren, Shaoqing, Kaiming He, Ross Girshick, and Jian Sun. "Faster r-cnn: Towards real-time object detection with region proposal networks." IEEE transactions on pattern analysis and machine intelligence 39, no. 6 (2016): 1137-1149. Chowdary, G. Jignesh, Narinder Singh Punn, Sanjay Kumar Sonbhadra, and Sonali Agarwal. "Face Mask Detection using Transfer Learning of InceptionV3." arXiv preprint arXiv:2009.08369 (2020). Jiang, Mingjie, and Xinqi Fan. "RetinaMask: A Face Mask detector." arXiv preprint arXiv:2005.03950 (2020).