imageInputLayer

Image input layer

Description

An image input layer inputs 2-D images to a network and applies data normalization.

For 3-D image input, use image3dInputLayer.

Creation

Syntax

layer = imageInputLayer(inputSize)

layer = imageInputLayer(inputSize,Name,Value)

Description

layer = imageInputLayer(inputSize) returns an image input layer and specifies the InputSize property.

layer = imageInputLayer(inputSize,Name,Value) sets the optional properties using name-value pairs. You can specify multiple name-value pairs. Enclose each property name in single quotes.

Properties

Image Input

InputSize — Size of the input

row vector of integers

Size of the input data, 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.

• For grayscale images, specify a vector with c equal to 1.

• For RGB images, specify a vector with c equal to 3.

• For multispectral or hyperspectral images, specify a vector with c equal to the number of channels.

For 3-D image or volume input, use image3dInputLayer.

Example: [224 224 3]

Normalization — Data normalization

'zerocenter' (default) | 'zscore' | 'rescale-symmetric' | 'rescale-zero-one' | 'none' | function handle

Data normalization to apply every time data is forward propagated through the input layer, specified as one of the following:

• 'zerocenter' — Subtract the mean specified by Mean.

• 'zscore' — Subtract the mean specified by Mean and divide by StandardDeviation.

• 'rescale-symmetric' — Rescale the input to be in the range [-1, 1] using the minimum and maximum values specified by Min and Max, respectively.

• 'rescale-zero-one' — Rescale the input to be in the range [0, 1] using the minimum and maximum values specified by Min and Max, respectively.

• 'none' — Do not normalize the input data.

• function handle — Normalize the data using the specified function. The function must be of the form Y = func(X), where X is the input data, and the output Y is the normalized data.

Tip

The software, by default, automatically calculates the normalization statistics at training time. To save time when training, specify the required statistics for normalization and set the 'ResetInputNormalization' option in trainingOptions to false.

NormalizationDimension — Normalization dimension

'auto' (default) | 'channel' | 'element' | 'all'

Normalization dimension, specified as one of the following:

• 'auto' – If the training option is false and you specify any of the normalization statistics (Mean, StandardDeviation, Min, or Max), then normalize over the dimensions matching the statistics. Otherwise, recalculate the statistics at training time and apply channel-wise normalization.

• 'channel' – Channel-wise normalization.

• 'element' – Element-wise normalization.

• 'all' – Normalize all values using scalar statistics.

Mean — Mean for zero-center and z-score normalization

[] (default) | 3-D array | numeric scalar

Mean for zero-center and z-score normalization, specified as a h-by-w-by-c array, a 1-by-1-by-c array of means per channel, a numeric scalar, or [], where h, w, and c correspond to the height, width, and the number of channels of the mean, respectively.

If you specify the Mean property, then Normalization must be 'zerocenter' or 'zscore'. If Mean is [], then the software calculates the mean at training time.

You can set this property when creating networks without training (for example, when assembling networks using assembleNetwork).

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

StandardDeviation — Standard deviation for z-score normalization

[] (default) | 3-D array | numeric scalar

Standard deviation for z-score normalization, specified as a h-by-w-by-c array, a 1-by-1-by-c array of means per channel, a numeric scalar, or [], where h, w, and c correspond to the height, width, and the number of channels of the standard deviation, respectively.

If you specify the StandardDeviation property, then Normalization must be 'zscore'. If StandardDeviation is [], then the software calculates the standard deviation at training time.

You can set this property when creating networks without training (for example, when assembling networks using assembleNetwork).

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

Min — Minimum value for rescaling

[] (default) | 3-D array | numeric scalar

Minimum value for rescaling, specified as a h-by-w-by-c array, a 1-by-1-by-c array of minima per channel, a numeric scalar, or [], where h, w, and c correspond to the height, width, and the number of channels of the minima, respectively.

If you specify the Min property, then Normalization must be 'rescale-symmetric' or 'rescale-zero-one'. If Min is [], then the software calculates the minimum at training time.

You can set this property when creating networks without training (for example, when assembling networks using assembleNetwork).

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

Max — Maximum value for rescaling

[] (default) | 3-D array | numeric scalar

Maximum value for rescaling, specified as a h-by-w-by-c array, a 1-by-1-by-c array of maxima per channel, a numeric scalar, or [], where h, w, and c correspond to the height, width, and the number of channels of the maxima, respectively.

If you specify the Max property, then Normalization must be 'rescale-symmetric' or 'rescale-zero-one'. If Max is [], then the software calculates the maximum at training time.

You can set this property when creating networks without training (for example, when assembling networks using assembleNetwork).

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

DataAugmentation — Data augmentation transforms

'none' (default) | 'randcrop' | 'randfliplr' | cell array of 'randcrop' and 'randfliplr'

Note

The DataAugmentation property is not recommended. To preprocess images with cropping, reflection, and other geometric transformations, use augmentedImageDatastore instead.

Data augmentation transforms to use during training, specified as one of the following.

• 'none' — No data augmentation

• 'randcrop' — Take a random crop from the training image. The random crop has the same size as the input size.

• 'randfliplr' — Randomly flip the input images horizontally with a 50% chance.

• Cell array of 'randcrop' and 'randfliplr'. The software applies the augmentation in the order specified in the cell array.

Augmentation of image data is another way of reducing overfitting [1], [2].

Data Types: string | char | cell

Layer

Name — Layer name

'' (default) | character vector | string scalar

Layer name, specified as a character vector or a string scalar. To include a layer in a layer graph, you must specify a nonempty unique layer name. If you train a series network with the layer and Name is set to '', then the software automatically assigns a name to the layer at training time.

Data Types: char | string

NumInputs — Number of inputs

0 (default)

Number of inputs of the layer. The layer has no inputs.

Data Types: double

InputNames — Input names

{} (default)

Input names of the layer. The layer has no inputs.

Data Types: cell

NumOutputs — Number of outputs

1 (default)

Number of outputs of the layer. This layer has a single output only.

Data Types: double

OutputNames — Output names

{'out'} (default)

Output names of the layer. This layer has a single output only.

Data Types: cell

Examples

Create Image Input Layer

Create an image input layer for 28-by-28 color images with name 'input'. By default, the layer performs data normalization by subtracting the mean image of the training set from every input image.

inputlayer = imageInputLayer([28 28 3],'Name','input')

inputlayer = 
  ImageInputLayer with properties:

                      Name: 'input'
                 InputSize: [28 28 3]

   Hyperparameters
          DataAugmentation: 'none'
             Normalization: 'zerocenter'
    NormalizationDimension: 'auto'
                      Mean: []

Include an image input layer in a Layer array.

layers = [ ...
    imageInputLayer([28 28 1])
    convolution2dLayer(5,20)
    reluLayer
    maxPooling2dLayer(2,'Stride',2)
    fullyConnectedLayer(10)
    softmaxLayer
    classificationLayer]

layers = 
  7x1 Layer array with layers:

     1   ''   Image Input             28x28x1 images with 'zerocenter' normalization
     2   ''   Convolution             20 5x5 convolutions with stride [1  1] and padding [0  0  0  0]
     3   ''   ReLU                    ReLU
     4   ''   Max Pooling             2x2 max pooling with stride [2  2] and padding [0  0  0  0]
     5   ''   Fully Connected         10 fully connected layer
     6   ''   Softmax                 softmax
     7   ''   Classification Output   crossentropyex

Compatibility Considerations

AverageImage property will be removed

Not recommended starting in R2019b

AverageImage will be removed. Use Mean instead. To update your code, replace all instances of AverageImage with Mean. There are no differences between the properties that require additional updates to your code.

imageInputLayer and image3dInputLayer, by default, use channel-wise normalization

Behavior change in future release

Starting in R2019b, imageInputLayer and image3dInputLayer, by default, use channel-wise normalization. In previous versions, these layers use element-wise normalization. To reproduce this behavior, set the NormalizationDimension option of these layers to 'element'.

References

[1] Krizhevsky, A., I. Sutskever, and G. E. Hinton. "ImageNet Classification with Deep Convolutional Neural Networks". Advances in Neural Information Processing Systems. Vol 25, 2012.

[2] Cireşan, D., U. Meier, J. Schmidhuber. "Multi-column Deep Neural Networks for Image Classification". IEEE Conference on Computer Vision and Pattern Recognition, 2012.

Extended Capabilities

GPU Code Generation
Generate CUDA® code for NVIDIA® GPUs using GPU Coder™.