function layer = imageInputLayer(varargin)

% imageInputLayer Image input layer

%

% layer = imageInputLayer(inputSize) defines an image input layer.

% inputSize is the size of the input images for the layer. It must be a

% row vector of two or three numbers.

%

% layer = imageInputLayer(inputSize, 'PARAM1', VAL1, 'PARAM2', VAL2, ...)

% specifies optional parameter name/value pairs for creating the layer:

%

% 'DataAugmentation' Specify data augmentations to use during training

% as a string or cell array of strings. Valid

% augmentations are 'randcrop', 'randfliplr', or

% 'none'.

%

% Default: 'none'

%

% 'Normalization' Specify the data normalization to apply as a

% string. Valid values are 'zerocenter' or 'none'.

% Normalization is applied every time data is

% forward propagated through the input layer.

%

% Default: 'zerocenter'

%

% 'Name' A name for the layer.

%

% Default: ''

%

% Example:

% Create an image input layer for 28-by-28 color images. At training

% time, images will be flipped from left to right with a probability

% of 50%.

%

% layer = imageInputLayer([28 28 3], ...

% 'DataAugmentation', 'randfliplr');

%

% See also nnet.cnn.layer.ImageInputLayer, convolution2dLayer,

% fullyConnectedLayer, maxPooling2dLayer.

% Copyright 2015-2016 The MathWorks, Inc.

% Parse the input arguments.

inputArguments = nnet.cnn.layer.ImageInputLayer.parseInputArguments(varargin{:});

normalization = iCreateTransforms(...

inputArguments.Normalization, inputArguments.InputSize);

augmentations = iCreateTransforms(...

inputArguments.DataAugmentation, inputArguments.InputSize);

% Create an internal representation of an image input layer.

internalLayer = nnet.internal.cnn.layer.ImageInput(...

inputArguments.Name, ...

inputArguments.InputSize, ...

normalization, ...

augmentations);

% Pass the internal layer to a function to construct a user visible image

% input layer.

layer = nnet.cnn.layer.ImageInputLayer(internalLayer);

end

function tformarray = iCreateTransforms(type, imageSize)

type = cellstr(type);

tformarray = nnet.internal.cnn.layer.ImageTransform.empty();

for i = 1:numel(type)

tnew = nnet.internal.cnn.layer.ImageTransformFactory.create(type{i}, imageSize);

tformarray = [tformarray tnew]; %#ok<AGROW>

end

end