图像分类网络

LeNet-5 1998

conv-5 max pooling 2*2 2
[CONV-POOL-CONV-POOL-CONV-FC]

AlexNet 2012

simplified AlexNet architecture.
[CONV-POOL-NORM-CONV-POOL-NORM-CONV-CONV-CONV-POOL-FC-FC-FC]

VGGNet 2014

only 3*3 CONV stride 1 pad 1 , and 2*2 MAX POOL stride 2

GoogleLeNet 2014

inception module.

ResNet 2015

Revolution of Depth 152 layer.
faster than a VGGNet. Even though it has 8x more layers.

Typical architectures look like
[(CONV-RELU)*N-POOL?]*(FC-RELU)*K, SOFTMAX.
ResNet/GoogleLeNet chanllenge this paradigm.

发布于: 2016年 03月 01日