In a custom CNN network, how can I know about the input size for each layer?

For preparing the type of table as shown in the above figure,how can I get values for the last column of the table i.e Input Size for each layer in my CNN model?

Answer 1

[Image Analyst]

When you build a custom network, for example like this:

layers = [
    imageInputLayer([227, 227, 3])
    convolution2dLayer(3,8,'Padding','same')
    batchNormalizationLayer
    reluLayer
    averagePooling2dLayer(2,'Stride',2)
    convolution2dLayer(3,16,'Padding','same')
    batchNormalizationLayer
    reluLayer
    averagePooling2dLayer(2,'Stride',2)
    convolution2dLayer(3,32,'Padding','same')
    batchNormalizationLayer
    reluLayer
    convolution2dLayer(3,32,'Padding','same')
    batchNormalizationLayer
    reluLayer
    dropoutLayer(0.2)
    fullyConnectedLayer(1)
    regressionLayer];

the 227 is not something you "get". It's something you specify. Bigger images take longer to train but will be more accurate. If your accuracy is low, try increasing the size of images you supply.

If you're doing transfer learning, like adapting alexnet or googlenet, then you need to know what image sizes they want and make sure you supply images of that size.

[debojit sharma]

No, I am not asking this question. My question is: I want to prepare a table for my project report like the one

given below

:

So, here in the last column of this table Input Size for each layer of the custom CNN network is shown. I am asking how can I know about the input size of each layer in my CNN network? Is there any function available in MatLab using which I can get these information?

[Ben]

You can get this information from analyzeNetwork. For example I can take @Image Analyst's network and do:

analyzeNetwork(layers)

The input size of a layer is equivalent to the activation size of the layers that connect into it. So for example the input size of conv_3 below is 56 x 56 x 16 as that's the activation size of the preceding layer:

[debojit sharma]

please tell me what will be the input size and output size of each layers in the following model.

layers = [

imageInputLayer([227, 227, 3])

convolution2dLayer(3,8,'Padding','same')

batchNormalizationLayer

reluLayer

averagePooling2dLayer(2,'Stride',2)

convolution2dLayer(3,16,'Padding','same')

batchNormalizationLayer

reluLayer

averagePooling2dLayer(2,'Stride',2)

convolution2dLayer(3,32,'Padding','same')

batchNormalizationLayer

reluLayer

convolution2dLayer(3,32,'Padding','same')

batchNormalizationLayer

reluLayer

flattenLayer('Name','flatten')

dropoutLayer(0.2)

lstmLayer(300,'OutputMode','last','Name','lstm')

fullyConnectedLayer(8)

softmaxLayer

classificationLayer];

[debojit sharma]

If i take my above example, is it so that output size of the lstmLayer is equal to the number of hidden units used in the lstmLayer?