http://www.cnblogs.com/nwpuxuezha/p/4302024.html
http://blog.csdn.net/cyh_24/article/details/51537709
注意:卷积时除不尽,按往小取整
池化时除不尽,按往大取整
如GoogleNet:
输入:100(batch_size) 3(三通道) 128*128
卷积:
convolution_param {
num_output: 128
pad: 3
kernel_size: 7
stride: 2
输出:100 128 64 64
池化:
pool: MAX
kernel_size: 3
stride: 2
输出:100 128 32 32
时间: 2024-10-11 16:31:23