Error using trainNetwork - Number of observations in X and Y disagree.

I want to build a cnn where my input is 1000 images of size 20 x 20 x 110 and output is 1000 images of size 20 x 20.

Therefore I have

net = trainNetwork(TB_train(:,:,:,:),TS_train(:,:,:),Layers,options); and I get an error 'Number of observations in X and Y disagree'. I don't understand as both of my matrices have 1000 images.

My cnn has these layers:

Layers = [imageInputLayer([20 20 110],'Normalization','none') %,'Weights',W,'Bias',B)

convolution2dLayer(3,128,'Padding','same')%,'Weights',W,'Bias',B)

batchNormalizationLayer

reluLayer

convolution2dLayer(1,64,'Padding','same')%,'Weights',W,'Bias',B)

batchNormalizationLayer

reluLayer

convolution2dLayer(1,32,'Padding','same')%,'Weights',W,'Bias',B)

batchNormalizationLayer

reluLayer

convolution2dLayer(1,16,'Padding','same')%,'Weights',W,'Bias',B)

batchNormalizationLayer

reluLayer

convolution2dLayer(1,1,'Padding','same')%,'Weights',W,'Bias',B)

batchNormalizationLayer

reluLayer

regressionLayer];

Answer 1

[Divya Gaddipati]

Try using

net = trainNetwork(TB_train, TS_train, Layers, options);

Answer 2

[Joseph Williams]

Hello, I know this is an old question, but I am having a very similar difficulty, except with 3D images. I have 4000 (4 x 256 x 4) 3-channel images and want to create 4000 (4 x 256 x 4) 1-channel images.

I have a very similar architecture to yours and using analyzeNetwork(layers) shows that the layer before regression puts in the right size (for both our networks).

What was your solution to generate 1000 images of size 20 x 20? it may be my solution as well.

Thanks!