Convolution 中 group 参数

最新推荐文章于 2024-10-12 18:29:57 发布

weixin_33737774

最新推荐文章于 2024-10-12 18:29:57 发布

阅读量160

点赞数

CC 4.0 BY-SA版权

文章标签：人工智能

原文链接：http://www.cnblogs.com/penguins/p/5103360.html

本文深入解析了Caffe官方文档与Alex Krizhevsky论文中关于group参数的定义及其在AlexNet中的应用，解释了group参数如何实现分组卷积，并讨论了其历史背景和在不同任务上的性能、速度与内存使用之间的权衡。

caffe 官方文档定义：

group (g) [default 1]: If g > 1, we restrict the connectivity of each filter to a subset of the input. Specifically, the input and output channels are separated into g groups, and the i-th output group channels will be only connected to the i-th input group channels.
Jia Yangqing 论坛回复：

It was there to implement the grouped convolution in Alex Krizhevsky's paper: when group=2, the first half of the filters are only connected to the first half of the input channels, and the second half only connected to the second half.

To group seems like a historical accident from the practical constraints of the time, but I don't know of a side-by-side comparison of group and no-group models. It could be nice to settle the question on the axes of task performance, speed + memory, and optimization time + hassle.
Alex Krizhevsky's paper 中conv的定义：

convolution_param {
num_output: 256
pad: 2
kernel_size: 5
group: 2
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
value: 1
}