How to share a fully connected layer and is it possible to use a channel direction pooling?

Question

Jin Yong Kim 2024-8-20

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2146589-how-to-share-a-fully-connected-layer-and-is-it-possible-to-use-a-channel-direction-pooling

评论： Jin Yong Kim 2024-8-21

(한국어로 답변주실 수 있다면 한국어로 답변 부탁드립니다.)

I am currently designing a model using the Deep Network Designer to implement a Convolutional Block Attention Module (CBAM).

For this, I need to use a channel-wise max/average pooling layer, and I am wondering if there is a way to implement this within MATLAB.

Additionally, I would like to know if there is a method to share a fully connected layer across different parts of the network. The current implementation of the fully connected layer seems to be limited to a single input and output.

Lastly, when passing through the fully connected layer, the dimension of the output is converted to (C x B). Is there a way to reshape this back to (S x S x C x B) in order to perform element-wise multiplication?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Angelo Yeo 2024-8-21

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2146589-how-to-share-a-fully-connected-layer-and-is-it-possible-to-use-a-channel-direction-pooling#answer_1502354

CBAM은 MATLAB에서 구현되어 있지 않습니다만, 가장 간단한 방법 중 하나는 Pytorch로 구현된 모델을 MATLAB에 불러와 사용하는 것입니다. 가령, 아래의 Repo에 들어있는 CBAM 모델을 pt 파일로 저장한 뒤에 MATLAB으로 가져와 사용할 수 있습니다.

External-Attention-pytorch/model/attention/CBAM.py at master · xmu-xiaoma666/External-Attention-pytorch (github.com)

Pytorch 모델을 MATLAB으로 가져올 때 아래와 같은 기능을 활용할 수 있습니다.

Import PyTorch network as MATLAB network - MATLAB importNetworkFromPyTorch (mathworks.com)