Input and output size for deep learning regression

Hi everyone,

I have the following input and target matrix

Input: 110 samples of 273x262

Target: 110 samples of 273x262

I have to work on deep learning regression problem with a simple layers as shown below

Layer: [imageInputLayer()    
    convolution2dLayer(5,16,'Padding','same')
    batchNormalizationLayer
    reluLayer
    fullyConnectedLayer()
    regressionLayer]

What is the matrix size I have to use for the inputlayer and fullyconnectedlayer?

I am thinking of 4D matrix of size [273, 262, 1, 110] for inputlayer and a 2D matrix of size [273*263, 110] for output layer.

Is this correct? Will this exceed the matrix array size preference? Any other suggestions. Thank you

Answer 1

[Uttiya Ghosh]

Hi SA,

From my understanding, you are working with grayscale images on a deep learning regression model. You are expecting a output in the form of a matrix for each image and not a single valued scalar output.

For imageInputLayer, size of the input data is specified as a row vector of integers [h w c], where h, w, and c correspond to the height, width, and number of channels respectively. You do not need to specify the number of samples. Hence, as per my understanding, the inputSize should be a row vector [273, 262, 1].

For fullyConnectedLayer, output size must be a positive integer. You shall not specify the sample size here as well. Hence as per my understanding, the outputSize should be 273*262.

For more information, refer to the following links.

https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.imageinputlayer.html

https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.fullyconnectedlayer.html