1. 训练过程、模型结构、数据的一些trick

2. 低精度、大batch; 1乘1卷积的stide; label smoothing和cosine learning rate decay

Bag of Freebies for Training Object Detection Neural Networks

SqueezeNet

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size

MobileNet

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

ShuffleNet

ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices

Xception

Xception: Deep Learning with Depthwise Separable Convolutions

inception

v1:

v2:

v3:

resnet

Deep Residual Learning for Image Recognition

DeepFace

DeepFace: Closing the Gap to Human-Level Performance in Face Verification

DeepID 系列

• Deep Learning Face Representation from Predicting 10,000 Classes [Yi Sun et al., 2014]

• Deep Learning Face Representation by Joint Identification-Verification [Yi Sun et al., 2014]

• Deeply learned face representations are sparse, selective, and robust [Yi Sun et al., 2014]

• DeepID3: Face Recognition with Very Deep Neural Networks [Yi Sun et al., 2015]

FaceNet(Triplet Loss)

FaceNet: A Unified Embedding for Face Recognition and Clustering [Florian Schroff et al., 2015]

SphereFace(A-Softmax)

• SphereFace: Deep Hypersphere Embedding for Face Recognition(A-Softmax loss) [Weiyang Liu al., 2017]

ArcFace

• ArcFace: Additive Angular Margin Loss for Deep Face Recognition [Jiankang Deng al., 2018]

Center loss

A Discriminative Feature Learning Approach for Deep Face Recognition [Yandong Wen et al., 2016]

L-Softmax

• Large-Margin Softmax Loss for Convolutional Neural Networks(L-Softmax loss) [Weiyang Liu al., 2017] code