Training failed: Layer 'classoutput' input size mismatch

I am training my convolutional neural network by using Deep Network Designer. When I did the training part it shows me: "Training failed: Layer 'classoutput' input size mismatch. Size of input to this layer is different from the expected input size. Inputs to this layers: from layer 'softmax' (output size 32x32x2)" My input contains 100000 sets of 32x32x2 data. Could anyone give me some hint to solve this problem, thank you!

Here is my network structure:

tempLayers = [

imageInputLayer([32 32 2],"Name","imageinput")

convolution2dLayer([9 9],256,"Name","conv_1_1","Padding","same")

averagePooling2dLayer([9 9],"Name","avgpool2d_1","Padding","same")

batchNormalizationLayer("Name","batchnorm_1_1")

leakyReluLayer(0.01,"Name","leakyrelu_1_1")

convolution2dLayer([5 5],256,"Name","conv_2_1","Padding","same")

averagePooling2dLayer([5 5],"Name","avgpool2d_2","Padding","same")

batchNormalizationLayer("Name","batchnorm_2_1")

leakyReluLayer(0.01,"Name","leakyrelu_2_1")

convolution2dLayer([5 5],256,"Name","conv_3_1","Padding","same")

averagePooling2dLayer([5 5],"Name","avgpool2d_3","Padding","same")

batchNormalizationLayer("Name","batchnorm_3_1")

leakyReluLayer(0.01,"Name","leakyrelu_3_1")

transposedConv2dLayer([5 5],2,"Name","transposed-conv_1","Cropping","same")

convolution2dLayer([5 5],256,"Name","conv_1_2","Padding","same")

batchNormalizationLayer("Name","batchnorm_1_2")

leakyReluLayer(0.01,"Name","leakyrelu_1_2")];

lgraph = addLayers(lgraph,tempLayers);

tempLayers = [

convolution2dLayer([5 5],256,"Name","conv_2_2","Padding","same")

batchNormalizationLayer("Name","batchnorm_2_2")

leakyReluLayer(0.01,"Name","leakyrelu_2_2")

convolution2dLayer([5 5],256,"Name","conv_3_2","Padding","same")

batchNormalizationLayer("Name","batchnorm_3_2")

leakyReluLayer(0.01,"Name","leakyrelu_3_2")

convolution2dLayer([5 5],256,"Name","conv_6","Padding","same")

batchNormalizationLayer("Name","batchnorm_5")

leakyReluLayer(0.01,"Name","leakyrelu_5")

convolution2dLayer([5 5],256,"Name","conv_7","Padding","same")

batchNormalizationLayer("Name","batchnorm_6")

leakyReluLayer(0.01,"Name","leakyrelu_6")];

lgraph = addLayers(lgraph,tempLayers);

tempLayers = [

additionLayer(2,"Name","addition")

transposedConv2dLayer([5 5],2,"Name","transposed-conv_2","Cropping","same")

convolution2dLayer([5 5],256,"Name","conv_4","Padding","same")

batchNormalizationLayer("Name","batchnorm_4")

leakyReluLayer(0.01,"Name","leakyrelu_4")

transposedConv2dLayer([5 5],4,"Name","transposed-conv_3","Cropping","same")

convolution2dLayer([9 9],2,"Name","conv_5","Padding","same")

softmaxLayer("Name","softmax")

classificationLayer("Name","output")];

lgraph = addLayers(lgraph,tempLayers);

clear tempLayers;

lgraph = connectLayers(lgraph,"leakyrelu_1_2","conv_2_2");

lgraph = connectLayers(lgraph,"leakyrelu_1_2","addition/in1");

lgraph = connectLayers(lgraph,"leakyrelu_6","addition/in2");

Answer 1

[Prateek Rai]

To my understanding, you are trying to train convolutional neural network by using Deep Network Designer but the input to 'classoutput' layer is mismatched.

The 'classoutput' layer is a classification layer and the input data to classification layer must have spatial dimension sizes equal to 1. You have to modify your neural network accordingly.

You can refer to classificationLayer MathWorks documentation page to find more on classification layer. You can also refer to analyzeNetwork MathWorks documentation page to analyze the deep learning network architecture.

[KAI-YANG WANG]

Thank you!