1.Insightface_pytorch版本的自定义数据准备过程
2.mxnet_insightface开源代码使用——自定义数据集制作
https://zhuanlan.zhihu.com/p/92738758
3.insight face数据打包(一):LFW
https://blog.csdn.net/ustczhng2012/article/details/113512500
https://github.com/deepinsight/insightface/blob/master/recognition/arcface_torch/docs/prepare_custom_dataset.md
https://github.com/deepinsight/insightface/blob/master/recognition/arcface_torch/docs/prepare_custom_dataset.md
6.AI 实战 - 基于 insightface 的人脸识别
https://blog.csdn.net/qq_25439881/article/details/125179117
7.face recognition based on yolov8 & Arcface
https://github.com/JOKER-3-z/EASY_USE_face_recognition
8.yoloV5-arcface_forlearn
https://github.com/ooooxianyu/yoloV5-arcface_forlearn
9. yoloV5-arcface_forlearn-master
10.Pytorch-FaceNet-DogDataset
FaceNet Implementation on DogFace Dataset
https://github.com/fcernafukuzaki/Pytorch-FaceNet-DogDataset
Paper To Code implementation of Facenet in native pytorch on Dog Dataset
11.insightface pytorch 答疑
https://blog.csdn.net/minstyrain/article/details/126901471
快速上手 PyTorch 人脸相似度计算方法 ArcFace
基于insightface实现的人脸识别和人脸注册-CSDN博客
https://yeyupiaoling.blog.csdn.net/article/details/108304894
人脸识别从原理到实践
https://github.com/zhongyy/SFace/tree/main/SFace_torch
https://github.com/deepinsight/insightface/wiki/Dataset-Zoo
- 训练公开数据
先从insightface datasets下载数据,里面提供了对齐(112x112)好的人脸文件,并打包成rec格式加速读取,如果需要原始文件图片,可参考load_images_from_bin.py自行保存,也可使用image_iter_rec.py里的FaceDataset直接加载数据
- insightface datasets:
- load_images_from_bin.py:
- image_iter_rec.py:
人脸识别0-10:insightFace-半监督系统搭建(5)-流程总结,代码分享。_insightface casia-facev5-CSDN博客
- 训练自己的数据
训练自己的数据要复杂很多,人半监督系统搭建(5)-流程总结首先通过爬虫等手段获取人脸图片,每个人放一个文件夹。https://github.com/DarkIsComing/1.FaceRecognition
MTCNN训练
要部署的话可以使用MTCNN,
https://github.com/imistyrain/MTCNN
其提供了部署全平台实现,包括C++、python、ncnn和tensorflow,还有加速版本和opencv直接加载版本,是所有版本中的集大成者.
如果想了解算法原理,可以参考
本文的训练流程参考至
提供了主流的pytorch实现
训练数据准备
MTCNN混合了三类任务: 人脸/非人脸分类、人脸框回归和人脸关键点定位. WIDER FACE (左图) 用来进行人脸分类和框回归. CNN_FacePoint (右图) 用来进行关键点定位.