#deeper

Going deeper with convolutions(GoogLeNet、Inception)

从LeNet-5开始,cnn就有了标准的结构:stackedconvolutionallayersarefollowedbyoneormorefully-connectedlayers。对于Imagenet这种大的数据集,趋势是增加层数和层的大小,用dropout解决过拟合。1×1卷积核在Inception中大量使用,...