Layer 4 is expected to have a different size(Convolutional Neural Network)?

Question

Esra Kaya 2017-4-3

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/333365-layer-4-is-expected-to-have-a-different-size-convolutional-neural-network

回答： Huma Chaudhry 2020-6-25

Hi, I am working on the segmentation of retina bloodvessels with CNN. While I compose my CNN structure, I come across a problem with the size of Layer 4 which is conv2 and can not train the network. My images have 625x625X1 gray images and my network structure is below. I appreciate your help already, thanks.

if true
  % code
end
  inputLayer = imageInputLayer([625 625 1]);
  conv1 = convolution2dLayer(25,25 , 'Stride', 25, 'Padding', 0, 'NumChannels', 1 );
  relu1 = reluLayer;
  conv2 = convolution2dLayer(4,22, 'Stride', 1, 'Padding', 0, 'NumChannels', 12);
  relu2 = reluLayer;
  norm2 = crossChannelNormalizationLayer(5);
  pool2 = maxPooling2dLayer(2, 'Stride', 2, 'Padding', 1); %subsampling 
  conv3 = convolution2dLayer(4,8 , 'Stride', 1, 'Padding', 0, 'NumChannels', 12);
  relu3 = reluLayer; 
  norm3 = crossChannelNormalizationLayer(5);
  pool3 = maxPooling2dLayer(2, 'Stride', 2, 'Padding', 1); %subsampling
  fc4 = fullyConnectedLayer(1100);
  relu4 = reluLayer;
  drop4 = dropoutLayer(0.5);
  fc5 = fullyConnectedLayer(1);
  SMAX = softmaxLayer;
  ClassOut = classificationLayer;

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

Huma Chaudhry 2020-6-25

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/333365-layer-4-is-expected-to-have-a-different-size-convolutional-neural-network#answer_456835

Most likely the issue is with the DEPTH of the block. Check the network parameters and see if the input network and output network have suitable depth.