人脸识别相关开源项目汇总

刺骨的言语ヽ痛彻心扉 2022-04-12 03:53 271阅读 0赞

博客：[blog.shinelee.me][] | [博客园][Link 1] | [CSDN][]

### 文章目录 ###

*  全任务
 *  人脸检测
 *  人脸识别
 *  参考

人脸识别流程包括人脸检测、人脸对齐、人脸识别等子任务，这里优先总结功能相对齐全的开源项目，再总结完成单个子任务的开源项目。本文主要关注方法较流行且提供源码的开源项目，忽略了仅提供SDK的。

# 全任务 #

1.  [SeetaFaceEngine][]与[SeetaFaceEngine2][]，是中科院山世光组开源的项目，含Detection、Alignment、Identification，前者代码齐全，很适合学习，只是没有开源模型训练方法，后者只提供了SDK。
2.  [OpenCV][]以及[opencv\_contrib][opencv_contrib]，传统方法和深度学习方法都有，4.0版本已发布，见[ChangeLog][]。
3.  [dlib][]基于C++的机器学习库，其中也含有 frontal face detection 和 face recognition等。[face\_recognition][face_recognition]基于dlib的人脸识别（python）。
4.  [openface官网][openface]以及[cmusatyalab/openface github][cmusatyalab_openface github]是基于google [FaceNet(CVPR 2015)][FaceNet_CVPR 2015]用Python和Torch实现的人脸识别系统。
5.  [facenet][]，基于Tensorflow实现的FaceNet，参考了上面的openface
6.  [TadasBaltrusaitis/OpenFace][TadasBaltrusaitis_OpenFace]（与上面的openface不是同一个），汇总了多篇论文的实现，含facial landmark detection、head pose estimation、facial action unit recognition 以及 eye-gaze estimation，侧重于facial behavior analysis and understanding。

# 人脸检测 #

1.  [MTCNN][]，Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Neural Networks。以及[MTCNN-light][]，无框架实现版本。
2.  百度冠军方法PyramidBox的两个非官方实现，[EricZgw/PyramidBox][EricZgw_PyramidBox]以及[Goingqs/PyramidBox][Goingqs_PyramidBox]，前者TensorFlow后者pytorch。

# 人脸识别 #

以下，copy自[大学生写人脸识别算法现实吗？ - 卡本特的回答 - 知乎][- _ -]

>  *  caffeFace：[https://github.com/ydwen/caffe-face][https_github.com_ydwen_caffe-face]
>  *  normFace：[https://github.com/happynear/NormFace][https_github.com_happynear_NormFace]
>  *  mobile ID：[liuziwei7/mobile-id][liuziwei7_mobile-id]
>  *  sphereFace：[wy1iu/sphereface][wy1iu_sphereface]
>  *  insightface：[deepinsight/insightface][deepinsight_insightface]

2018年初Insightface曾拿Megaface第一，参见[InsightFace - 使用篇, 如何一键刷分LFW 99.80%, MegaFace 98%][InsightFace - _ _LFW 99.80_ MegaFace 98]。

以上，待更新。

# 参考 #

*  [https://www.zhihu.com/question/64860792][https_www.zhihu.com_question_64860792]
 *  [计算机视觉人脸相关开源项目总结][Link 2]

[blog.shinelee.me]: https://blog.shinelee.me/
[Link 1]: https://www.cnblogs.com/shine-lee/
[CSDN]: https://blog.csdn.net/blogshinelee
[SeetaFaceEngine]: https://github.com/seetaface/SeetaFaceEngine
[SeetaFaceEngine2]: https://github.com/seetaface/SeetaFaceEngine2
[OpenCV]: https://github.com/opencv/opencv
[opencv_contrib]: https://github.com/opencv/opencv_contrib
[ChangeLog]: https://github.com/opencv/opencv/wiki/ChangeLog
[dlib]: http://dlib.net/
[face_recognition]: https://github.com/ageitgey/face_recognition#face-recognition
[openface]: http://cmusatyalab.github.io/openface/
[cmusatyalab_openface github]: https://github.com/cmusatyalab/openface
[FaceNet_CVPR 2015]: http://www.cv-foundation.org/openaccess/content_cvpr_2015/app/1A_089.pdf
[facenet]: https://github.com/davidsandberg/facenet
[TadasBaltrusaitis_OpenFace]: https://github.com/TadasBaltrusaitis/OpenFace
[MTCNN]: https://github.com/kpzhang93/MTCNN_face_detection_alignment
[MTCNN-light]: https://github.com/AlphaQi/MTCNN-light
[EricZgw_PyramidBox]: https://github.com/EricZgw/PyramidBox
[Goingqs_PyramidBox]: https://github.com/Goingqs/PyramidBox
[- _ -]: https://www.zhihu.com/question/64860792/answer/233782977
[https_github.com_ydwen_caffe-face]: https://github.com/ydwen/caffe-face
[https_github.com_happynear_NormFace]: https://github.com/happynear/NormFace
[liuziwei7_mobile-id]: https://github.com/liuziwei7/mobile-id
[wy1iu_sphereface]: https://github.com/wy1iu/sphereface
[deepinsight_insightface]: https://github.com/deepinsight/insightface
[InsightFace - _ _LFW 99.80_ MegaFace 98]: https://zhuanlan.zhihu.com/p/33750684
[https_www.zhihu.com_question_64860792]: https://www.zhihu.com/question/64860792
[Link 2]: https://blog.csdn.net/chaipp0607/article/details/78885720