图像分类--模型汇总

image classification

model 简介 conv_padding conv_activation pooling
LeNet5 3个卷积层+2个pooling层 valid sigmoid
AlexNet 8层 same ReLU Overlapping Pooling
VGG-19 19层 same ReLU
GoogLeNet 22层
ResNet 152层

conv2d的VALID方式不会在原有输入的基础上添加新的像素