caffe下自己模型的搭建

论文中有这么一段描述:
Due to the small input size, no spatial pooling or downsampling is performed.
Starting from the first layer, the input image is convolved with 96 filters of size
9 × 9, resulting in a map of size 16 × 16 (to avoid boundary effects) and 96
channels. The 96 channels are then pooled with maxout in group of size g = 2,
resulting in 48 channels.
图片1.jpg

其中The 96 channels are then pooled with maxout in group of size g = 2,
resulting in 48 channels.对应虚线圆圈中的图本人比较困惑,请问要怎么实现呢?是下图的方法实现吗??
P70527-182534.jpg


 
 
 
 
 
 
 
 
 
 
 
 
 

3 个评论

是的,先slice再eltwise取max
请问Slice和Eltwise层caffe下都有自带吧
有,百度下怎么用就可以

要回复文章请先登录注册