FaceBoxes in PyTorch

By Zisian Wong, Shifeng Zhang

A PyTorch implementation of FaceBoxes: A CPU Real-time Face Detector with High Accuracy. The official code in Caffe can be found here.

Performance

| Dataset | Original Caffe | PyTorch Implementation | |:-|:-:|:-:| | AFW | 98.98 % | 98.47% | | PASCAL | 96.77 % | 96.84% | | FDDB | 95.90 % | 95.44% |

Citation

Please cite the paper in your publications if it helps your research:

@inproceedings{zhang2017faceboxes,
  title = {Faceboxes: A CPU Real-time Face Detector with High Accuracy},
  author = {Zhang, Shifeng and Zhu, Xiangyu and Lei, Zhen and Shi, Hailin and Wang, Xiaobo and Li, Stan Z.},
  booktitle = {IJCB},
  year = {2017}
}

Contents

• Installation

• Training

• Evaluation

• References

Installation

1. Install PyTorch >= v1.0.0 following official instruction.

2. Clone this repository. We will call the cloned directory as $FaceBoxes_ROOT.

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

1. Compile the nms:

./make.sh

_Note: Codes are based on Python 3+._

Training

1. Download WIDER FACE dataset, place the images under this directory: Shell $FaceBoxes_ROOT/data/WIDER_FACE/images

2. Convert WIDER FACE annotations to VOC format or download our converted annotations, place them under this directory:

   $FaceBoxes_ROOT/data/WIDER_FACE/annotations

3. Train the model using WIDER FACE:

   cd $FaceBoxes_ROOT/
   python3 train.py

If you do not wish to train the model, you can download our pre-trained model and save it in $FaceBoxes_ROOT/weights.

Evaluation

1. Download the images of AFW, PASCAL Face and FDDB to:

$FaceBoxes_ROOT/data/AFW/images/$FaceBoxes_ROOT/data/PASCAL/images/$FaceBoxes_ROOT/data/FDDB/images/

1. Evaluate the trained model using:

# dataset choices = ['AFW', 'PASCAL', 'FDDB']python3 test.py --dataset FDDB# evaluate using cpupython3 test.py --cpu

1. Download eval_tool to evaluate the performance.

References

• Official release (Caffe)

• A huge thank you to SSD ports in PyTorch that have been helpful:

  
  

  _Note: If you can not download the converted annotations, the provided images and the trained model through the above links, you can download them through BaiduYun._

• ssd.pytorch, RFBNet