Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

GCM--Identity mapping #1

Open
Klaviersonate opened this issue Dec 3, 2020 · 5 comments
Open

GCM--Identity mapping #1

Klaviersonate opened this issue Dec 3, 2020 · 5 comments

Comments

@Klaviersonate
Copy link

GCM模块的恒等映射分支，看了结构图和代码，请教下您，这里是先用了一个卷积再是non-local结构吗？卷积核尺寸是什么，在此处有什么作用呢？non-local的输出通道根据什么决定的呢？
@ReaFly
Copy link
Owner

ReaFly commented Dec 3, 2020 •
edited
Loading

您好，卷积核尺寸是1*1的，作用是降维，non-local模块的输出通道数与输入是一致的

@Klaviersonate
Copy link
Author

您好，卷积核尺寸是1*1的，作用是降维，non-local模块的输出通道数与输入是一致的

non-local有bottleneck，也是降维，就是降维再升维减少参数量。那么1*1卷积的降维和non-local的降维有什么区别吗

@ReaFly
Copy link
Owner

ReaFly commented Dec 3, 2020

1*1的降维是为了和其他三个分支的输出通道数保持一致，即经降维得到一个更为压缩的特征后再借助non-local捕获长距离依赖关系，non-local内部的降维再升维是降低该模块的计算量

@Klaviersonate
Copy link
Author

您好，称为 恒等映射，是non-local包含一个短连接，某种角度看non-local和resnet引入短连接类似，都是 输出H(x)=F(x)+输入x， 这个意思吗

@ReaFly
Copy link
Owner

ReaFly commented Dec 6, 2020

恒等映射更多的还是相对于其他三个分支所说的，因为该分支没有做pooling操作，只是借助non-local捕获长距离依赖关系

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@Klaviersonate @ReaFly