Caiyuan-Zheng / Real-time-face-recognition

using yolo-v3 mobilefacenet to recognite faces and estimate age and gender

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Real-time-face-recognition

using yolo-v3 and mobilefacenet to recognite faces and estimate age and gender

reference

https://github.com/yeziyang1992/Face_Recognition_Client
https://github.com/sirius-ai/MobileFaceNet_TF\n
https://github.com/ninesky110/Real-time-face-recognition
https://github.com/gittigxuy/yolo-v3_face_detection

environment

python==3.5.3
tensorflow==1.9.0
cuda==9.0
single RTX 2080Ti

Detection

一开始使用的mtcnn模型做检测,但是由于mtcnn在检测人脸的过程中会做关键点检测,随着检测到的人脸数的增加,检测所需要的时间也会增加。 后来改用ssd模型,ssd模型速度足够快(10ms),但是检测精度不高。 最后使用yolo-v3模型,精度相比ssd有了很大的提升,速度也在能接受的范围之内(在50ms左右)。

Recognition

尝试过facenet和mobilefacenet,因为两个模型在lfw数据集都已经得到很高的正确率,于是我采用了速度快不少的mobilefacenet

Run

下载项目到本地 git clone https://github.com/1029127253/Real-time-face-recognition.git
cd model-weights
下载模型文件,链接:https://docs.google.com/uc?export=download&id=1a_pbXPYNj7_Gi6OxUqNo_T23Dt_9CzOV
unzip YOLO_Face.h5.zip
cd ..
python main-9.py

About

using yolo-v3 mobilefacenet to recognite faces and estimate age and gender


Languages

Language:Python 99.3%Language:Shell 0.7%