Tensorflow.js tf.layers.separableConv2d() Function

Tensorflow.js is an open-source library developed by Google for running machine learning models and deep learning neural networks in the browser or node environment. It also helps the developers to develop ML models in JavaScript language and can use ML directly in the browser or in Node.js.

The tf.layers.separableCov2d() function is separable convolution which is a way to factorize a convolution kernel into two smaller kernels. Separable convolution consists of depthwise convolution and pointwise convolution which mixes together the resulting output channels. The depthMultiplier argument controls how many output channels are generated per input channel in the depthwise step. separableCov2d is Depthwise separable 2D convolution. 

Syntax: 

tf.layers.separableCov2d( args )

Parameters:

• args: It accepts objects as parameters with the following fields:
  • depthMultiplier: It is the number of depthwise convolution output channels for each input channel. It is equal to filtersIn * depthMultiplier. 
  • depthwiseInitializer: It is the initializer for the depthwise kernel matrix.
  • pointwiseInitializer: It is the initializer for the pointwise kernel matrix.
  • depthwiseRegularizer: It is a regularizer function applied to the depthwise kernel matrix.
  • pointwiseRegularizer: It is a regularizer function applied to the pointwise kernel matrix.
  • depthwiseConstraint: It is a constraint function applied to the depthwise kernel matrix.
  • pointwiseConstraint: It is a constraint function applied to the pointwise kernel matrix.
  • filters: It is the dimensionality of the output space.
  • kernelSize: It is a dimension that the convolution window will have.
  • strides: It is strides of convolution layer in each dimension. 
  • padding: It should be 'valid', 'same', and 'causal'. It defines the padding mode. 
  • dataFormat: It defines the format of data, which tells the ordering of the dimensions in the inputs.
  • dilationRate: The dilation rate to which the convolution layer should dilate in each dimension.
  • activation: It is the Activation function of the layer. If you don't specify the activation, none is applied.
  • useBias: It defines whether the layer uses a bias vector. Default is true.
  • kernelInitializer: It is the initializer for the convolution kernel weight matrix.
  • biasInitializer: It is the initializer for the bias vector. 
  • kernelConstraint: It is a constraint for the convolution kernel weights.
  • biasConstraint: It is a constraint for the bias vector.
  • kernelRegularizer: It is a regularizer function applied to the kernel weight matrix.
  • biasRegularizer: It is a regularizer function applied to the bias vector.
  • activityRegularizer: It is a regularizer function applied to the activation.
  • inputShape: It should be an array of numbers. This field is used to create an input layer which is used to be inserted before this layer.
  • batchInputShape: It should be an array of numbers. This field will be used if inputShape and this field are provided as a parameter for creating the input layer which is used to insert before this layer.
  • batchSize: It should be a number. In the absence of batchInputShape this field is used to create batchInputShape with inputShape. batchInputShape : [ batchSize ,  ...inputShape].
  • dtype: If this layer is used as the input layer, then this field is used as the data type for this layer.
  • name: It should be a string type. this field defines the name for this layer.
  • trainable: It should be boolean. This field defines whether the weights of this layer are trainable with fit or not.
  • weights: This should be a tensor that defines the initial weight value for this layer.
  • inputDType: This is a data type that is used for Legacy support.

Returns: It returns SeparableConv2D.

Example 1:

import * as tf from "@tensorflow/tfjs";

// InputShape and Input layer for convLstm2dCell layer
const InputShape = [ 4,  5, 2];
const input = tf.input({ shape: InputShape });

// Creating ConvLstm2dCell 
const separableConv2d = tf.layers.separableConv2d(
    { filters: 3, kernelSize: 2, batchInputShape: [ 4, 5, 3]});
const output = separableConv2d.apply(input);

// Printing summary of layers
const model = tf.model({ inputs: input, outputs: output });
model.summary();

Output:

__________________________________________________________________________________________
Layer (type)                Input Shape               Output shape              Param #   
==========================================================================================
input12 (InputLayer)        [[null,4,5,2]]            [null,4,5,2]              0         
__________________________________________________________________________________________
separable_conv2d_SeparableC [[null,4,5,2]]            [null,3,4,3]              17        
==========================================================================================
Total params: 17
Trainable params: 17
Non-trainable params: 0
__________________________________________________________________________________________

Example 2:

// Import the header file
import * as tf from "@tensorflow/tfjs"

// Creating separableConv2d layer
const separableConv2d = tf.layers.separableConv2d({ 
    filters : 2, 
    kernelSize: 3,
    batchInputShape: [2, 3, 5, 5]
});

// Create an input with 2 time steps.
const input = tf.input({shape: [3, 4, 5]});
const output = separableConv2d.apply(input);

// Printing the Shape of file
console.log(JSON.stringify(output.shape));

Output:

[null,1,2,2]

Reference: https://js.tensorflow.org/api/latest/#layers.separableConv2d