help error CNN The output of layer 6 is incompatible with the input expected by layer 7.

hi:

I am using MatLab R2017b in Win 10 64 bits

Greetings, can anybody help me with this code

Code:

     % Define a CNN architecture
      conv1 = convolution2dLayer(200,96,'Padding',2,'BiasLearnRateFactor',2);
      conv1.Weights = gpuArray(single(randn([200 200 3 96])*0.0001));

      fc1 = fullyConnectedLayer(225,'BiasLearnRateFactor',2);
      fc1.Weights = gpuArray(single(randn([225 1130])*0.1));

     fc2 = fullyConnectedLayer(10,'BiasLearnRateFactor',2);
     fc2.Weights = gpuArray(single(randn([10 225])*0.1));

     layers = [ ...
               imageInputLayer([480 720 3]);
               conv1;
               maxPooling2dLayer(3,'Stride',2);
               reluLayer();
               convolution2dLayer(100,256,'Padding',2,'BiasLearnRateFactor',2);
               reluLayer();
               averagePooling2dLayer(85,'Stride',2);
               convolution2dLayer(75,384,'Padding',2,'BiasLearnRateFactor',2);
               reluLa yer();
               averagePooling2dLayer(3,'Stride',2);
               fc1;
               reluLayer();
               fc2;
               softmaxLayer()
               classificationLayer()    
           ];

          % Define the training options.

           opts = trainingOptions('sgdm', ...
                                  'InitialLearnRate', 0.001, ...
                                  'LearnRateSchedule', 'piecewise', ...
                                  'LearnRateDropFactor', 0.1, ...
                                 'LearnRateDropPeriod', 8, ...
                                 'L2Regularization', 0.004, ...
                                 'MaxEpochs', 10, ...
                                 'MiniBatchSize', 100, ...
                                 'Verbose', true);

          %Training the CNN

          [net, info] = trainNetwork(XTrain, TTrain, layers, opts);

the error is:

" Error using trainNetwork (line 140)
The output of layer 6 is incompatible with the input expected by layer 7.

Error in CNN2 (line 87)
[net, info] = trainNetwork(XTrain, TTrain, layers, opts);

Caused by:
    Error using nnet.internal.cnn.layer.util.inferParameters>iInferSize (line 86)
    The output of layer 6 is incompatible with the input expected by layer 7.
"

The error is when run the code: trainNetwork

thks in advanced for help

Best regarts

Angel lerma

Answer 1

[Joss Knight]

I ran the Network Analyzer on your network:

analyzeNetwork(layers)

The resulting report told me that the activations input to your average pooling layer 7 are too small for a pooling window of 85-by-85 (see attached).

Did you really mean to make an average pooling layer with such a large window? because that would be unusual. Actually, it's unusual that you wanted a 200-by-200 convolution kernel followed by a 100-by-100 kernel, since these are almost the size of the original images.