“Identity Mappings in Deep Residual Networks", 2016

Inception-V1(2014) Inception-V2 = Inception-V1 + BN(batch normalization) Inception-V3(2015) = nception-V2 + onvolution factorization + label smoothing + Auxiliary classifier Inception-V4(2016) = Inception-ResNet

추후 다시 살펴 보기 : 라온피플 블로그Class49, Class50

• 코드(Keras) : https://github.com/myutwo150/keras-inception-resnet-v2

Inception-V4

1. 개요

기존 ResNet에 pre-activation 개념 적용

2. 특징

2.1 Residual Network Module

• f(x) : activation(ReLU) 함수
• h(x) : identity 함수
• F(x) : Residual 함수