Error when using lstm with cnn

2 次查看（过去 30 天）
Mohammed Firas
Mohammed Firas 2024-7-19
回答： Kaustab Pal 2024-8-20
XTrain = single(DL_input_reshaped(:,1,1,Training_Ind));
YTrain = single(DL_output_reshaped(1,1,:,Training_Ind)); XValidation = single(DL_input_reshaped(:,1,1,Validation_Ind));
YValidation = single(DL_output_reshaped(1,1,:,Validation_Ind));
YValidation_un = single(DL_output_reshaped_un);
%% DL Model definition with adjusted pooling and convolution layers layers = [ imageInputLayer([size(XTrain,1), 1, 1],'Name','input','Normalization','none')
convolution2dLayer(3, 64, 'Padding', 'same', 'Name', 'conv1')
batchNormalizationLayer('Name', 'bn1')
reluLayer('Name', 'relu1')
maxPooling2dLayer([3,1], 'Stride', [3,1], 'Name', 'maxpool1')
convolution2dLayer(3, 128, 'Padding', 'same', 'Name', 'conv2')
batchNormalizationLayer('Name', 'bn2')
reluLayer('Name', 'relu2')
maxPooling2dLayer([3,1], 'Stride', [3,1], 'Name', 'maxpool2')
convolution2dLayer(3, 256, 'Padding', 'same', 'Name', 'conv3')
batchNormalizationLayer('Name', 'bn3')
reluLayer('Name', 'relu3')
maxPooling2dLayer([3,1], 'Stride', [3,1], 'Name', 'maxpool3')
flattenLayer('Name', 'flatten') % Flatten to 1D per sample
lstmLayer(200, 'OutputMode', 'last', 'Name', 'lstm1') % LSTM layer
fullyConnectedLayer(512, 'Name', 'fc1')
reluLayer('Name', 'relu4')
dropoutLayer(0.5, 'Name', 'dropout1')
fullyConnectedLayer(1024, 'Name', 'fc2')
reluLayer('Name', 'relu5')
dropoutLayer(0.5, 'Name', 'dropout2')
fullyConnectedLayer(2048, 'Name', 'fc3')
reluLayer('Name', 'relu6')
dropoutLayer(0.5, 'Name', 'dropout3')
fullyConnectedLayer(size(YTrain,3), 'Name', 'fc4')
regressionLayer('Name', 'output') ];
options = trainingOptions('rmsprop', ...
.
.
.
so this error is appear to me
((error useing trainNetwork Invalid training data.
The output size (1024) of the last layer does not match the response size (1).))
so the size or XTrain and YTrain is (features x 1 x 1 x minbatchsize)
  1 个评论
Walter Roberson
Walter Roberson 2024-7-19
XTrain = single(DL_input_reshaped(:,1,1,Training_Ind));
You are training with (something by 1 by 1 by something-else) data.
The networks probably expect (something by something-else) -- 2D data instead of 4D data.

请先登录，再进行评论。

请先登录，再回答此问题。

回答（1 个）

Kaustab Pal
Kaustab Pal 2024-8-20
Hi @Mohammed Firas
It appears you're facing an issue due to a size mismatch between the output of your final fully connected layer and the expected response size in YTrain. Specifically, the network's final layer is generating an output of size 1024, while the expected size is 1.
To resolve this, please ensure that YTrain is appropriately structured for regression tasks. It should have dimensions [1, 1, numResponses, miniBatchSize]. If numResponses is 1, the final fully connected layer should produce a single output value per sample.
To correct this, modify the final layer named "fc4" to output one value per sample by using the following configuration:
fullyConnectedLayer(1, 'Name', 'fc4')
I hope this resolves the issue.

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

R2023a

Community Treasure Hunt

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

Start Hunting!

Translated by