- Use models provided by biubug, centerface, dlib, linzaer, mobileface, mtcnn, zqcnn, pig to detect faces.
- Use dilb, pfld, pig, zqcnn models or the two L106Net models to detect key points of the face (ranging from 68-106 points)
- Support the detection of pictures, videos and camera input
- Support the visualization and export of test results
- Support recording inference time
- windows10
- python3.6
- numpy==1.20.2
- opencv-python==4.5.1.48
- dlib==19.6.1(dlib)
- mtcnn==0.1.0(mtcnn)
- pytorch==1.4.0(biubug)
- onnx==1.9.0(linzaer)
- onnxruntime==1.8.1(linzaer, pfld, centerface, frda, pig)
- mxnet==1.5.0(mobileface, zqcnn)
- tensorflow==1.5.0(cnn)
The project contains all pre-trained models that may be used, just clone them directly
# From your favorite development directory:
git clone https://github.com/MickeyLLG/face_det_lib.git
Video files, camera serial numbers or image files need to be fed in as parameters. If nothing is fed in, it will detect test/test_15fps.avi as default
Any video format supported by OpenCV is available (mp4
, avi
etc.):
python demo.py --video /path/to/video.mp4 --save_path /path/to/save.mp4 --face_det fd --landmark_det ld
Need to declare the serial number of the camera used:
python demo.py --cam 0 --save_path /path/to/save.mp4 --face_det fd --landmark_det ld
Any image format supported by OpenCV is available (jpg
, jpeg
etc.):
python demo.py --image /path/to/image.jpg --save_path /path/to/save.jpg --face_det fd --landmark_det ld
--face_det
parameter determines the method used for face detection, and the optional methods include(pig
is recommended for better performance):
dlib
,mtcnn
,linzaer
,centerface
,biubug
,mobileface
,zqmtcnn
,pig
--landmark_det
parameter determines the method used for face key point detection. The optional methods include(pig
is recommended for better performance):
dlib
,pfld
,L106Net112
,L106Net96
,cnn
,frda
,pig
- 使用biubug、centerface、dlib、linzaer、mobileface、mtcnn、zqcnn提供的模型来检测人脸
- 使用dilb、pfld模型和zqcnn提供的两种L106Net模型来检测面部关键点(68-106点不等)
- 支持对图片、视频以及摄像头输入的检测
- 支持检测结果可视化并导出
- 支持记录推理时间
- windows10
- python3.6
- numpy==1.20.2
- opencv-python==4.5.1.48
- dlib==19.6.1(dlib)
- mtcnn==0.1.0(mtcnn)
- pytorch==1.4.0(biubug)
- onnx==1.9.0(linzaer)
- onnxruntime==1.8.1(linzaer, pfld, centerface, frda, pig)
- mxnet==1.5.0(mobileface, zqcnn)
- tensorflow==1.5.0(cnn)
项目内包含所有可能用到的预训练模型,直接克隆即可
# From your favorite development directory:
git clone https://github.com/MickeyLLG/face_det_lib.git
视频文件、摄像头序号或图像文件需要作为参数传入。如果没有传入,则会默认检测test/test_15fps.avi
OpenCV支持的任何视频格式均可 (mp4
, avi
etc.):
python demo.py --video /path/to/video.mp4 --save_path /path/to/save.mp4 --face_det fd --landmark_det ld
需要声明使用的摄像头序号:
python demo.py --cam 0 --save_path /path/to/save.mp4 --face_det fd --landmark_det ld
OpenCV支持的任何图片格式均可 (jpg
, jpeg
etc.),路径中不要有中文:
python demo.py --image /path/to/image.jpg --save_path /path/to/save.jpg --face_det fd --landmark_det ld
--face_det
参数决定了人脸检测使用的方法,可选方法包括(推荐pig
,效果较好):
dlib
,mtcnn
,linzaer
,centerface
,biubug
,mobileface
,zqmtcnn
,pig
--landmark_det
参数决定了人脸关键点检测使用的方法,可选方法包括(推荐pig
,效果较好):
dlib
,pfld
,L106Net112
,L106Net96
,cnn
,frda
,pig