how to copy layers and connections from an existing neural network?

6 次查看（过去 30 天）
Jack Xiao
Jack Xiao 2018-7-11
编辑： Jack Xiao 2018-7-12
just installed the latest version and find the support for deep learning is better and better. here, i have a question. I want to create a new network, but i do not want to write from scratch. take googlenet for example, I want to create a new network which can be a little like googlenet. but not all the same. maybe I need to refer some layers or structure. so how to copy layers and connections from an existing neural network?

请先登录，再回答此问题。

回答（1 个）

Von Duesenberg
Von Duesenberg 2018-7-11
Something along the lines (I had an exemple withe Alexnet, but the basic principle should be identical; here, I just resize the input layers because I have gray, not RGB, images, and I have 45 classes):
net = alexnet;
layers = net.Layers;
layers(1) = imageInputLayer([227, 227,1]);
layers(2) = convolution2dLayer([11, 11], 96,'Padding',0, 'Stride', 4, 'BiasLearnRateFactor',2);
layers(23) = fullyConnectedLayer(45);
layers(24) = softmaxLayer();
layers(25) = classificationLayer();
  1 个评论
Jack Xiao
Jack Xiao 2018-7-12
编辑：Jack Xiao 2018-7-12
thanks, I make a mistake previously. my copy operation makes a struct not a Layer. maybe it is a common operation in matlab. take " layersKept = net.Layers(1:25) ; " for example, it is for copying. and for connection copy: connectionKept = net.Connections(1:27,1:2) ;

请先登录，再进行评论。

类别

在 Help Center 和 File Exchange 中查找有关 Image Data Workflows 的更多信息

标签

产品


版本

R2018a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by