roiAlignLayer

Non-quantized ROI pooling layer for Mask-CNN

Since R2020b

Description

An ROI align layer outputs fixed size feature maps for every rectangular ROI within an input feature map. Use this layer to create a Mask R-CNN network.

Given an input feature map of size [H W C N], where C is the number of channels and N is the number of observations, the output feature map size is [h w C sum(M)], where h and w are the specified output size. M is a vector of length N and M(i) is the number of ROIs associated with the i-th input feature map.

There are two inputs to this layer:

• 'in' — The input feature map

• 'roi' — A list of ROIs to pool

Use the input names when connecting or disconnecting the ROI align layer to other layers using connectLayers (Deep Learning Toolbox) or disconnectLayers (Deep Learning Toolbox) (requires Deep Learning Toolbox™).

Creation

Syntax

layer = roiAlignLayer(outputSize)

layer = roiAlignLayer(outputSize,Name,Value)

Description

layer = roiAlignLayer(outputSize) creates an ROI align layer with pooled output size outputSize. The outputSize input sets the OutputSize property.

layer = roiAlignLayer(outputSize,Name,Value) set properties of the ROI align layer by using one or more name-value pair arguments. Enclose each property name in quotes.

For example, roiAlignLayer([7 7],'Name','roialignlayer') creates an ROI align layer with a pooled output size of 7-by-7 pixels and name 'roialignlayer'.

Properties

OutputSize — Pooled output size

vector of two positive integers

Pooled output size, specified as a vector of two positive integers [h w], where h is the height and w is the width.

Data Types: double

ROIScale — Scale of input feature map to input image

1 (default) | positive number

Scale of the input feature map to the input image, specified as a positive number.

Data Types: double

SamplingRatio — Number of samples in each pooled bin

'auto' (default) | row vector of two positive integers

Number of samples in each pooled bin, specified as 'auto' or a row vector of two positive integers. The two elements are the number of vertical and horizontal samples, respectively.

If you do not specify the sampling ratio, then the number of vertical samples has the default value ceil(roiHeight/outputHeight). Likewise, the number of horizontal samples has the default value ceil(roiWidth/outputWidth).

Data Types: double | char

Name — Layer name

"" (default) | character vector | string scalar

Layer name, specified as a character vector or a string scalar. For Layer array input, the trainnet (Deep Learning Toolbox) and dlnetwork (Deep Learning Toolbox) functions automatically assign names to layers with the name "".

The ROIAlignLayer object stores this property as a character vector.

Data Types: char | string

NumInputs — Number of inputs

2 (default)

Number of inputs of the layer. This layer accepts two inputs.

Data Types: double

InputNames — Input names

{'in' 'roi'} (default)

Input names of the layer.

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

Create ROI Align Layer

Specify the pooled output size.

outputSize = [7 7];

Create an ROI align layer named 'roialign'.

layer = roiAlignLayer(outputSize,'Name','roialign')

layer = 
  ROIAlignLayer with properties:

             Name: 'roialign'
        NumInputs: 2
       InputNames: {'in'  'roi'}
       OutputSize: [7 7]

   Hyperparameters
         ROIScale: 1
    SamplingRatio: 'auto'

More About

ROI Align Layer

An ROI align layer outputs fixed size feature maps for every rectangular ROI within an input feature map. The layer first partitions an ROI into fixed sized bins of size OutputSize without quantizing the grid points. Each bin is further sampled at SamplingRatio locations. The value at each sampled point is inferred using bilinear interpolation. The average of the sampled values is returned as the output value of each pooled bin.

Version History

Introduced in R2020b