Training for libfacedetection in PyTorch

It is the training program for libfacedetection. The source code is based on MMDetection. Some data processing functions from SCRFD modifications.

Visualization of our network architecture: [netron].

Contents

• Installation
• Preparation
• Training
• Detection
• Evaluation on WIDER Face
• Export CPP source code
• Export to ONNX model
• Design your own model
• Citation

Installation

1. Install PyTorch >= v1.7.0 following official instruction. e.g.\ On GPU platforms (cu102):\ shell conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.2 -c pytorch
2. Install MMCV >= v1.3.17 following official instruction. e.g.\ shell pip install mmcv-full==1.3.17 -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.7.0/index.html
3. Clone this repository. We will call the cloned directory as $TRAIN_ROOT. Shell git clone https://github.com/Wwupup/wwfacedet cd wwfacedet python setup.py develop
4. Install dependencies. shell pip install -r requirements.txt

Note: Codes are based on Python 3+.

Preparation

1. Download the WIDER Face dataset and its evaluation tools.
2. Extract zip files under $TRAIN_ROOT/data/widerface as follows: shell $ tree data/widerface data/widerface ├── wider_face_split ├── WIDER_test ├── WIDER_train ├── WIDER_val └── labelv2 ├── train │ └── labelv2.txt └── val ├── gt └── labelv2.txt

Training

Following MMdetection training processing.

Shell CUDA_VISIBLE_DEVICES=0,1 bash tools/dist_train.sh ./config/yunet/x.py 2

Detection

Shell python tools/detect-image.py ./config/x.py ./work_dirs/x/latest.pth ./image.jpg

Evaluation on WIDER Face

shell python tools/test_widerface.py ./config/x.py ./work_dirs/x/latest.pth --mode 2

Performance on WIDER Face (Val): confidencethreshold=0.02, nmsthreshold=0.45, in origin size:

AP_easy=0.899, AP_medium=0.883, AP_hard=0.792

Export CPP source code

The following bash code can export a CPP file for project libfacedetection

Shell python tools/export2cpp.py ./config/x.py ./work_dirs/x/latest.pth

Export to onnx model

Export to onnx model for libfacedetection/example/opencv_dnn.

shell python tools/wwdet2onnx.py ./config/x.py ./work_dirs/x/latest.pth

Compare ONNX model with other works

Inference on exported ONNX models using ONNXRuntime:

shell python tools/compare_inference.py ./onnx/wwdet.onnx --mode AUTO --eval --score_thresh 0.02 --nms_thresh 0.45

Some similar approaches(e.g. SCRFD, Yolo5face, retinaface) to inference are also supported.

With Intel i7-12700K and input_size = origin size, score_thresh = 0.3, nms_thresh = 0.45, some results are list as follow:

| Model | APeasy | APmedium | AP_hard | #Params | Params Ratio | MFlops | Forward (ms) | | ----------------------- | ------- | --------- | ------- | ------- | ------------ | ------ | ------------ | | SCRFD0.5(ICLR2022) | 0.879 | 0.863 | 0.759 | 631410 | 7.43x | 184 | 22.3 | | Retinaface0.5(CVPR2020) | 0.899 | 0.866 | 0.660 | 426608 | 5.02X | 245 | 13.9 | | YuNet(Ours) | 0.885 | 0.877 | 0.762 | 85006 | 1.0x | 136 | 10.6 |

The compared ONNX model is available in https://share.weiyun.com/nEsVgJ2v Password：gydjjs

Citation

The loss used in training is EIoU, a novel extended IoU. More details can be found in:

@article{eiou, author={Peng, Hanyang and Yu, Shiqi}, journal={IEEE Transactions on Image Processing}, title={A Systematic IoU-Related Method: Beyond Simplified Regression for Better Localization}, year={2021}, volume={30}, pages={5032-5044}, doi={10.1109/TIP.2021.3077144} }

The paper can be open accessed at https://ieeexplore.ieee.org/document/9429909.

We also published a paper on face detection to evaluate different methods.

@article{facedetect-yu, author={Yuantao Feng and Shiqi Yu and Hanyang Peng and Yan-ran Li and Jianguo Zhang} title={Detect Faces Efficiently: A Survey and Evaluations}, journal={IEEE Transactions on Biometrics, Behavior, and Identity Science}, year={2021} }

The paper can be open accessed at https://ieeexplore.ieee.org/document/9580485