标签:ide output filter info 常用 channels 输入 png 指定
卷积神经网络组成: INPUT -CONV -ReLU -POOL -FC
卷积操作:
输出:
Input: rows * rows * channels, N filter_rows*filter_rows filters with stride=stride, pade=pad
Output: output_rows = (rows+2*pad-filter_rows)/stride + 1
MAX POOLING更常用,AVE POOLING用得少了
标签:ide output filter info 常用 channels 输入 png 指定
原文地址:https://www.cnblogs.com/xbit/p/9563442.html