FaceMaskDetection-pytorch

face mask detection, including face detection and mask recognition

1.dataset

Here we mainly use dataset RMFD and Baidu open dataset. In total 14000s for training, and 3500s for testing.

RMFD dataset url: https://github.com/X-zhangyang/Real-World-Masked-Face-Dataset

2.training

Running script:

sh train.sh

Model is efficient and high-accuracy, with real-time speed on the CPU devices.

	Parameters	Accuracy
MobileNetv3 large	12M	99.0%
MobileNetv3 small	3.8M	98.6%

model download:

链接： https://pan.baidu.com/s/1fP9q4nELXm6dQQirbPy6cA 提取码：kjkx

examples:

3.Deploy

We use ncnn to deploy it on Android devices.

convert pytorch model to onnx model:

cd tools
python pytorch2onnx.py

# simpllifier
python -m onnxsim face_mask.onnx face_mask_sim.onnx

convert onnx to ncnn:

./onnx2ncnn face_mask_sim.onnx face_mask_sim.param face_mask_sim.bin

4.References

https://github.com/zisianw/FaceBoxes.PyTorch

https://github.com/kuan-wang/pytorch-mobilenet-v3

https://github.com/X-zhangyang/Real-World-Masked-Face-Dataset