• 设为首页
  • 点击收藏
  • 手机版
    手机扫一扫访问
    迪恩网络手机版
  • 关注官方公众号
    微信扫一扫关注
    迪恩网络公众号

xuexingyu24/MobileFaceNet_Tutorial_Pytorch: This repo illustrates how to impleme ...

原作者: [db:作者] 来自: 网络 收藏 邀请

开源软件名称(OpenSource Name):

xuexingyu24/MobileFaceNet_Tutorial_Pytorch

开源软件地址(OpenSource Url):

https://github.com/xuexingyu24/MobileFaceNet_Tutorial_Pytorch

开源编程语言(OpenSource Language):

Jupyter Notebook 92.5%

开源软件介绍(OpenSource Introduction):

MobileFaceNet_Tutorial_Pytorch

  • This repo illustrates how to implement MobileFaceNet and Arcface for face recognition task.
  • Pretrained model is posted for tests over picture, video and cam
  • Help document on how to implement MTCNN+MobileFaceNet is available
  • Scripts on transforming MXNET data records in Insightface to images are provided
  • Scripts on train and evaluation of MobileFaceNet model are provided

MobileFaceNet Video Demo

Test over Picture, Video and Cam

  1. Test Picture
python MTCNN_MobileFaceNet.py -img {image_path}
  1. Take Picture for Face Database
  • over cam
    python take_picture.py -n {name}
    
  • over photo
    python take_ID.py -i {image_path} -n {name}
    
  1. Test Video
  • over cam
    python cam_demo.py
    
  • over video file
    python video_demo.py
    
  1. Instruction
MobileFaceNet_Step_by_Step.ipynb

Train

Download training and evaluation data from Model Zoo. All training data has been cropped, aligned and resized as 112 x 112. Put images and annotation files into "data_set" folder. The structure should be arranged as follows:

data_set/
            ---> AgeDB-30
            ---> CASIA_Webface_Image
            ---> CFP-FP
            ---> faces_emore_images
            ---> LFW
  1. The following script is provided to convert .bin and .rec file to images:
python data_set/load_images_from_bin.py
  1. Generate the corresponding annotation files
python data_set/anno_generation.py
  1. Train MobileFaceNet
python Train.py
  1. Instruction
MobileFaceNet_Training_Step_by_Step.ipynb

The training results over faces_emore data (5822653 images / 85742 ids) are shown below:

Evaluation

python Evaluation.py

Here is the evaluation result. 'Flip' the image could be applied to encode the embedding feature vector with ~ 0.07% higer accuracy. L2 distance score slightly outperforms cos similarity (not necessarily the same trend for other cases, but it is what we conclude in this work)

Eval Type Score LFW AgeDB-30 CFP-FP
Flip L2 99.52 96.30 92.93
Flip Cos 99.50 96.18 92.84
UnFlip L2 99.45 95.63 93.10
UnFlip Cos 99.45 95.65 93.10

Don't forget to star the repo if it is helpful for your research

Reference




鲜花

握手

雷人

路过

鸡蛋
该文章已有0人参与评论

请发表评论

全部评论

专题导读
热门推荐
阅读排行榜

扫描微信二维码

查看手机版网站

随时了解更新最新资讯

139-2527-9053

在线客服(服务时间 9:00~18:00)

在线QQ客服
地址:深圳市南山区西丽大学城创智工业园
电邮:jeky_zhao#qq.com
移动电话:139-2527-9053

Powered by 互联科技 X3.4© 2001-2213 极客世界.|Sitemap