GroupedConvolution2DLayer - 2-D grouped convolutional layer - MATLAB (original) (raw)

2-D grouped convolutional layer

Description

A 2-D grouped convolutional layer separates the input channels into groups and applies sliding convolutional filters. Use grouped convolutional layers for channel-wise separable (also known as depth-wise separable) convolution.

For each group, the layer convolves the input by moving the filters along the input vertically and horizontally and computing the dot product of the weights and the input, and then adding a bias term. The layer combines the convolutions for each group independently. If the number of groups is equal to the number of channels, then this layer performs channel-wise convolution.

Creation

Syntax

Description

`layer` = groupedConvolution2dLayer(`filterSize`,`numFiltersPerGroup`,'channel-wise') creates a layer for channel-wise convolution (also known as depth-wise convolution). In this case, the software determines the NumGroups property at training time. This syntax is equivalent to setting NumGroups to the number of input channels.

example

`layer` = groupedConvolution2dLayer(___,[Name,Value](#namevaluepairarguments)) sets the optional Stride, DilationFactor, Parameters and Initialization, Learning Rate and Regularization, andName properties using name-value pairs. To specify input padding, use the 'Padding' name-value pair argument. For example,groupedConvolution2dLayer(5,128,2,'Padding','same') creates a 2-D grouped convolutional layer with 2 groups of 128 filters of size[5 5] and pads the input to so that the output has the same size. You can specify multiple name-value pairs. Enclose each property name in single quotes.

example

Input Arguments

expand all

Name-Value Arguments

Use comma-separated name-value pair arguments to specify the size of the padding to add along the edges of the layer input or to set the Stride, DilationFactor, Parameters and Initialization, Learning Rate and Regularization, andName properties. Enclose names in single quotes.

Example: groupedConvolution2dLayer(5,128,2,'Padding','same') creates a 2-D grouped convolutional layer with 2 groups of 128 filters of size[5 5] and pads the input to so that the output has the same size.

Padding — Input edge padding

[0 0 0 0] (default) | vector of nonnegative integers | "same"

Input edge padding, specified as one of these values:

• "same" — Add padding of size calculated by the software at training or prediction time so that the output has the same size as the input when the stride equals 1. If the stride is larger than 1, then the output size isceil(inputSize/stride), where inputSize is the height or width of the input and stride is the stride in the corresponding dimension. The software adds the same amount of padding to the top and bottom, and to the left and right, if possible. If the padding that must be added vertically has an odd value, then the software adds extra padding to the bottom. If the padding that must be added horizontally has an odd value, then the software adds extra padding to the right.
• Nonnegative integer p — Add padding of sizep to all the edges of the input.
• Vector [a b] of nonnegative integers — Add padding of sizea to the top and bottom of the input and padding of sizeb to the left and right.
• Vector [t b l r] of nonnegative integers — Add padding of size t to the top, b to the bottom,l to the left, and r to the right of the input.

Example: Padding=1 adds one row of padding to the top and bottom, and one column of padding to the left and right of the input.

Example: Padding="same" adds padding so that the output has the same size as the input (if the stride equals 1).

Properties

expand all

Grouped Convolution

FilterSize — Height and width of filters

vector of two positive integers

Height and width of the filters, specified as a vector [h w] of two positive integers, where h is the height and w is the width. FilterSize defines the size of the local regions to which the neurons connect in the input.

When you create the layer, you can specify FilterSize as a scalar to use the same value for the height and width.

Example: [5 5] specifies filters with a height of 5 and a width of 5.

NumFiltersPerGroup — Number of filters per group

positive integer

Number of filters per group, specified as a positive integer. This property determines the number of channels in the output of the layer. The number of output channels is FiltersPerGroup * NumGroups.

Example: 10

NumGroups — Number of groups

positive integer | 'channel-wise'

Number of groups, specified as a positive integer or'channel-wise'.

If NumGroups is'channel-wise', then the software creates a layer for channel-wise convolution (also known as depth-wise convolution). In this case, the layer determines the NumGroups property at training time. This value is equivalent to settingNumGroups to the number of input channels.

The number of groups must evenly divide the number of channels of the layer input.

Example: 2

Stride — Step size for traversing input

[1 1] (default) | vector of two positive integers

Step size for traversing the input vertically and horizontally, specified as a vector[a b] of two positive integers, where a is the vertical step size and b is the horizontal step size. When creating the layer, you can specify Stride as a scalar to use the same value for both step sizes.

Example: [2 3] specifies a vertical step size of 2 and a horizontal step size of 3.

DilationFactor — Factor for dilated convolution

[1 1] (default) | vector of two positive integers

Factor for dilated convolution (also known as atrous convolution), specified as a vector [h w] of two positive integers, where h is the vertical dilation and w is the horizontal dilation. When creating the layer, you can specify DilationFactor as a scalar to use the same value for both horizontal and vertical dilations.

Use dilated convolutions to increase the receptive field (the area of the input which the layer can see) of the layer without increasing the number of parameters or computation.

The layer expands the filters by inserting zeros between each filter element. The dilation factor determines the step size for sampling the input or equivalently the upsampling factor of the filter. It corresponds to an effective filter size of (Filter Size – 1) .* Dilation Factor + 1. For example, a 3-by-3 filter with the dilation factor [2 2] is equivalent to a 5-by-5 filter with zeros between the elements.

Example: [2 3]

PaddingSize — Size of padding

[0 0 0 0] (default) | vector of four nonnegative integers

Size of padding to apply to input borders, specified as a vector[t b l r] of four nonnegative integers, where t is the padding applied to the top, b is the padding applied to the bottom, l is the padding applied to the left, and r is the padding applied to the right.

When you create a layer, use the 'Padding' name-value pair argument to specify the padding size.

Example: [1 1 2 2] adds one row of padding to the top and bottom, and two columns of padding to the left and right of the input.

PaddingMode — Method to determine padding size

"manual" (default) | "same"

Method to determine padding size, specified as "manual" or"same".

The software automatically sets the value of PaddingMode based on the Padding value you specify when creating a layer.

• If you set the Padding option to a scalar or a vector of nonnegative integers, then the software automatically sets PaddingMode to"manual".
• If you set the Padding option to"same", then the software automatically setsPaddingMode to'same' and calculates the size of the padding at training time so that the output has the same size as the input when the stride equals 1. If the stride is larger than 1, then the output size isceil(inputSize/stride), where inputSize is the height or width of the input and stride is the stride in the corresponding dimension. The software adds the same amount of padding to the top and bottom, and to the left and right, if possible. If the padding that must be added vertically has an odd value, then the software adds extra padding to the bottom. If the padding that must be added horizontally has an odd value, then the software adds extra padding to the right.

PaddingValue — Value to pad data

0 (default) | scalar | "symmetric-include-edge" | "symmetric-exclude-edge" | "replicate"

Value to pad data, specified as one of these values:

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

NumChannelsPerGroup — Number of channels per group

'auto' (default) | positive integer

Number of channels per group, specified as 'auto' or a positive integer. The number of channels per group is equal to the number of input channels divided by the number of groups.

The software automatically sets this property at training time.

Example: 256

Parameters and Initialization

WeightsInitializer — Function to initialize weights

'glorot' (default) | 'he' | 'narrow-normal' | 'zeros' | 'ones' | function handle

Function to initialize the weights, specified as one of the following:

• 'glorot' – Initialize the weights with the Glorot initializer [1] (also known as Xavier initializer). The Glorot initializer independently samples from a uniform distribution with zero mean and variance 2/(numIn + numOut), where numIn = FilterSize(1)*FilterSize(2)*NumChannelsPerGroup and numOut = FilterSize(1)*FilterSize(2)*NumFiltersPerGroup.
• 'he' – Initialize the weights with the He initializer [2]. The He initializer samples from a normal distribution with zero mean and variance 2/numIn, wherenumIn = FilterSize(1)*FilterSize(2)*NumChannelsPerGroup.
• 'narrow-normal' – Initialize the weights by independently sampling from a normal distribution with zero mean and standard deviation 0.01.
• 'zeros' – Initialize the weights with zeros.
• 'ones' – Initialize the weights with ones.
• Function handle – Initialize the weights with a custom function. If you specify a function handle, then the function must be of the form weights = func(sz), where sz is the size of the weights. For an example, see Specify Custom Weight Initialization Function.

The layer only initializes the weights when theWeights property is empty.

Data Types: char | string | function_handle

BiasInitializer — Function to initialize biases

"zeros" (default) | "narrow-normal" | "ones" | function handle

Function to initialize the biases, specified as one of these values:

• "zeros" — Initialize the biases with zeros.
• "ones" — Initialize the biases with ones.
• "narrow-normal" — Initialize the biases by independently sampling from a normal distribution with a mean of zero and a standard deviation of 0.01.
• Function handle — Initialize the biases with a custom function. If you specify a function handle, then the function must have the formbias = func(sz), where sz is the size of the biases.

The layer initializes the biases only when the Bias property is empty.

The GroupedConvolution2DLayer object stores this property as a character vector or a function handle.

Data Types: char | string | function_handle

Weights — Layer weights

[] (default) | numeric array

Layer weights for the layer, specified as a numeric array.

The layer weights are learnable parameters. You can specify the initial value of the weights directly using the Weights property of the layer. When you train a network, if the Weights property of the layer is nonempty, then the trainnet function uses the Weights property as the initial value. If the Weights property is empty, then the software uses the initializer specified by the WeightsInitializer property of the layer.

At training time, Weights is aFilterSize(1)-by-FilterSize(2)-by-NumChannelsPerGroup-by-NumFiltersPerGroup-by-NumGroups array, where NumInputChannels is the number of channels of the layer input.

Data Types: single | double

Bias — Layer biases

[] (default) | numeric array

Layer biases for the layer, specified as a numeric array.

The layer biases are learnable parameters. When you train a neural network, if Bias is nonempty, then the trainnet function uses the Bias property as the initial value. IfBias is empty, then software uses the initializer specified by BiasInitializer.

At training time, Bias is a 1-by-1-by-NumFiltersPerGroup-by-NumGroups array.

Data Types: single | double

Learning Rate and Regularization

WeightLearnRateFactor — Learning rate factor for weights

1 (default) | nonnegative scalar

Learning rate factor for the weights, specified as a nonnegative scalar.

The software multiplies this factor by the global learning rate to determine the learning rate for the weights in this layer. For example, if WeightLearnRateFactor is 2, then the learning rate for the weights in this layer is twice the current global learning rate. The software determines the global learning rate based on the settings you specify using the trainingOptions function.

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

BiasLearnRateFactor — Learning rate factor for biases

1 (default) | nonnegative scalar

Learning rate factor for the biases, specified as a nonnegative scalar.

The software multiplies this factor by the global learning rate to determine the learning rate for the biases in this layer. For example, if BiasLearnRateFactor is 2, then the learning rate for the biases in the layer is twice the current global learning rate. The software determines the global learning rate based on the settings you specify using the trainingOptions function.

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

L2 regularization factor for the weights, specified as a nonnegative scalar.

The software multiplies this factor by the global L2 regularization factor to determine the L2 regularization for the weights in this layer. For example, if WeightL2Factor is 2, then the L2 regularization for the weights in this layer is twice the global L2 regularization factor. You can specify the global L2 regularization factor using the trainingOptions function.

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

BiasL2Factor — L2 regularization factor for biases

0 (default) | nonnegative scalar

L2 regularization factor for the biases, specified as a nonnegative scalar.

The software multiplies this factor by the global L2 regularization factor to determine the L2 regularization for the biases in this layer. For example, if BiasL2Factor is 2, then the L2 regularization for the biases in this layer is twice the global L2 regularization factor. The software determines the global L2 regularization factor based on the settings you specify using the trainingOptions function.

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

Layer

Layer name, specified as a character vector or string scalar. For Layer array input, the trainnet anddlnetwork functions automatically assign names to layers with the name "".

The GroupedConvolution2DLayer object stores this property as a character vector.

Data Types: char | string

NumInputs — Number of inputs

1 (default)

This property is read-only.

Number of inputs to the layer, returned as 1. This layer accepts a single input only.

Data Types: double

InputNames — Input names

{'in'} (default)

This property is read-only.

Input names, returned as {'in'}. This layer accepts a single input only.

Data Types: cell

NumOutputs — Number of outputs

1 (default)

This property is read-only.

Number of outputs from the layer, returned as 1. This layer has a single output only.

Data Types: double

OutputNames — Output names

{'out'} (default)

This property is read-only.

Output names, returned as {'out'}. This layer has a single output only.

Data Types: cell

Examples

collapse all

Create Grouped Convolution Layer

Create a grouped convolutional layer with 3 groups of 10 filters, each with a height and width of 11, and the name 'gconv1'.

layer = groupedConvolution2dLayer(11,10,3,'Name','gconv1')

layer = GroupedConvolution2DLayer with properties:

               Name: 'gconv1'

Hyperparameters FilterSize: [11 11] NumGroups: 3 NumChannelsPerGroup: 'auto' NumFiltersPerGroup: 10 Stride: [1 1] DilationFactor: [1 1] PaddingMode: 'manual' PaddingSize: [0 0 0 0] PaddingValue: 0

Learnable Parameters Weights: [] Bias: []

Use properties method to see a list of all properties.

Create Channel-Wise Convolution Layer

Create a channel-wise convolutional (also known as depth-wise convolutional) layer with groups of 10 filters, each with a height and width of 11, and the name 'cwconv1'.

layer = groupedConvolution2dLayer(11,10,'channel-wise','Name','cwconv1')

layer = GroupedConvolution2DLayer with properties:

               Name: 'cwconv1'

Hyperparameters FilterSize: [11 11] NumGroups: 'channel-wise' NumChannelsPerGroup: 'auto' NumFiltersPerGroup: 10 Stride: [1 1] DilationFactor: [1 1] PaddingMode: 'manual' PaddingSize: [0 0 0 0] PaddingValue: 0

Learnable Parameters Weights: [] Bias: []

Use properties method to see a list of all properties.

Create Layers for Channel-Wise Separable Convolution

A typical convolutional neural network contains blocks of convolution, batch normalization, and ReLU layers. For example,

filterSize = 3; numFilters = 16;

convLayers = [ convolution2dLayer(filterSize,numFilters,'Stride',2,'Padding','same') batchNormalizationLayer reluLayer];

For channel-wise separable convolution (also known as depth-wise separable convolution), replace the convolution block with channel-wise convolution and point-wise convolution blocks.

Specify the filter size and the stride in the channel-wise convolution and the number of filters in the point-wise convolution. For the channel-wise convolution, specify one filter per group. For point-wise convolution, specify filters of size 1 in convolution2dLayer.

cwsConvLayers = [ groupedConvolution2dLayer(filterSize,1,'channel-wise','Stride',2,'Padding','same') batchNormalizationLayer reluLayer

convolution2dLayer(1,numFilters,'Padding','same')
batchNormalizationLayer
reluLayer];

Create a network containing layers for channel-wise separable convolution.

layers = [ imageInputLayer([227 227 3])

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

groupedConvolution2dLayer(3,1,'channel-wise','Stride',2,'Padding','same')
batchNormalizationLayer
reluLayer
convolution2dLayer(1,16,'Padding','same')
batchNormalizationLayer
reluLayer

maxPooling2dLayer(2,'Stride',2)

fullyConnectedLayer(5)
softmaxLayer];

Algorithms

expand all

Layer Input and Output Formats

Layers in a layer array or layer graph pass data to subsequent layers as formatted dlarray objects. The format of a dlarray object is a string of characters in which each character describes the corresponding dimension of the data. The formats consist of one or more of these characters:

• "S" — Spatial
• "C" — Channel
• "B" — Batch
• "T" — Time
• "U" — Unspecified

For example, you can describe 2-D image data that is represented as a 4-D array, where the first two dimensions correspond to the spatial dimensions of the images, the third dimension corresponds to the channels of the images, and the fourth dimension corresponds to the batch dimension, as having the format "SSCB" (spatial, spatial, channel, batch).

You can interact with these dlarray objects in automatic differentiation workflows, such as those for developing a custom layer, using a functionLayer object, or using the forward and predict functions withdlnetwork objects.

This table shows the supported input formats of GroupedConvolution2DLayer objects and the corresponding output format. If the software passes the output of the layer to a custom layer that does not inherit from the nnet.layer.Formattable class, or aFunctionLayer object with the Formattable property set to 0 (false), then the layer receives an unformatted dlarray object with dimensions ordered according to the formats in this table. The formats listed here are only a subset. The layer may support additional formats such as formats with additional "S" (spatial) or"U" (unspecified) dimensions.

References

[1] Glorot, Xavier, and Yoshua Bengio. "Understanding the Difficulty of Training Deep Feedforward Neural Networks." In Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, 249–356. Sardinia, Italy: AISTATS, 2010. https://proceedings.mlr.press/v9/glorot10a/glorot10a.pdf

[2] He, Kaiming, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. "Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification." In 2015 IEEE International Conference on Computer Vision (ICCV), 1026–34. Santiago, Chile: IEEE, 2015. https://doi.org/10.1109/ICCV.2015.123

Extended Capabilities

C/C++ Code Generation

Generate C and C++ code using MATLAB® Coder™.

Usage notes and limitations:

• Code generation for the ARM Compute Library is not supported for a 2-D grouped convolution layer that has the NumGroups property set to an integer value greater than two.
• For code generation, the PaddingValue parameter must be equal to 0, which is the default value.
• Code generation does not support passingdlarray objects with unspecified (U) dimensions to this layer.

GPU Code Generation

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

Usage notes and limitations:

• Code generation for the ARM Mali GPU is not supported for a 2-D grouped convolution layer that has the NumGroups property set as 'channel-wise' or a value greater than two.
• For code generation, the PaddingValue parameter must be equal to 0, which is the default value.
• Code generation does not support passingdlarray objects with unspecified (U) dimensions to this layer.

Version History

Introduced in R2019a