ResNet
https://blog.csdn.net/weixin_43624538/article/details/85049699
CNN
https://my.oschina.net/u/876354/blog/1634322
YoloV2
https://blog.csdn.net/shanlepu6038/article/details/84778770
https://blog.csdn.net/yanhx1204/article/details/81017134
YoloV3
http://www.pianshen.com/article/3460315699/
BN
https://www.cnblogs.com/guoyaohua/p/8724433.html
GoogleNet
Traditionally, the most direct way to improve deep neural networks is to increase the size of the network , including width and depth. Depth refers to the number of layers in the network, and width refers to the number of neurons used in each layer. However, there are two major disadvantages in this simple and direct solution :
(1) An increase in network size also means an increase in parameters, which makes the network easier to overfit .
(2) An increase in computing resources .
The author proposes that the fundamental solution is to transform the fully connected structure into a sparsely connected structure .
https://blog.csdn.net/weixin_43624538/article/details/84863685
https://www.jianshu.com/p/22e3af789f4e
Inception v2&v3
https://blog.csdn.net/sunbaigui/article/details/50807418
https://blog.csdn.net/weixin_43624538/article/details/84963116
From Inception v1, v2, v3, v4, RexNeXt to Xception to MobileNets, ShuffleNet, MobileNetV2, ShuffleNetV2, MobileNetV3
https://blog.csdn.net/qq_14845119/article/details/73648100
Deep ID series
https://blog.csdn.net/weixin_42546496/article/details/88537882
CenterLoss
https://www.aiuai.cn/aifarm102.html
L-Softmax loss
https://blog.csdn.net/u014380165/article/details/76864572
A-Softmax loss
https://blog.csdn.net/weixin_42546496/article/details/88062272 (translation of the paper)
FaceNet
https://blog.csdn.net/qq_15192373/article/details/78490726 (Paper explanation)
https://blog.csdn.net/ppllo_o/article/details/90707295 (paper translation)
ArcFace/IncightFace
https://blog.csdn.net/weixin_42546496/article/details/88387325
https://blog.csdn.net/hanjiangxue_wei/article/details/86368948 (Explanation of loss function code)