In this repository, we provide training data, network settings and loss designs for deep face detection. The training data is Widerface datasets, which were already packed in darknet format.he network backbone is darknet53, classifical loss is focal loss,which improve 1.2% Map on Widerface val.read focal loss and code,please check the blog.
This repository can help researcher/engineer to develop deep face detection algorithms quickly by only two steps: download the dataset and run the training script.In sum,we can run a face detection system in window easily.
All Widerface data are into VOC format.Please click BaiduDrive 提取码：r4wc
- if you want to make data into Voc format with yourself, please check
data_process/widerfaceToVoc.ipynb
Requirements
-
On windows
- Visual Studio 2015 or 2017
- cuda 10.0
- cudnn 7
- openCV3.0
-
On linux
- cuda9.0
- openCV3.0
- cudnn7
1、Clone the repository
git clone https://github.com/jmu201521121021/FaceDetector-Base-Yolov3-spp.git
2、Install darknetAB with GPU surpport
- detail,please check here
3、Download the training set (Widerface) and place it in darknet/build/darknet/x64/data/voc
4、go to darknet/build/darknet/x64/data/voc/
python voc_label.py
5、start train,into ./train
sh train.sh
- Please check here BaiduDrive 提取码：if0r
-
Test on widerFace val dataset
-
Test on FDDB
- Requiremets
- windows10
- GPU memory >=3GB (Nvidia version)
- Download,Please check BaiduDrive 提取码：zfik .
[Chen Jinshan](jinshanchen[at]jmu.edu.cn)