Welcome to our Support Center
Updated

SeparableConv2D

Description

Define the separable conv 2D layer according to its parameters. To be used for the TimeDistributed layer. Type : polymorphic.

 

Input parameters

 

Β parameters :Β layer parameters.

Β n_filters :Β integer, the dimensionality of the output space.
Default value β€œ3”.
sizeΒ : array integer, Β specify the height and width of the 2D convolution window. Can be a single integer to specify the same value for all spatial dimensions.
Default value “[3,3]”.
stride : array integer, specify the strides of the convolution along the height and width. Can be a single integer to specify the same value for all spatial dimensions. Β Current implementation only supports equal length strides in the row and column dimensions.
Default value “[1,1]”.
Β activation :Β enum, activation function to use.
Default value β€œrelu”.
Β optimizer :

Β algorithm :Β enum, (name of optimizer) for optimizer instance.
Default value β€œadam”.
Β learning_rate :Β float, define the learning rate to use.
Default value β€œ0.001”.
Β beta_1 :Β float, define the exponential decay rate for the 1st moment estimates.
Default value β€œ0.9”.
Β beta_2 :Β float, define the exponential decay rate for the 2nd moment estimates.
Default value β€œ0.999”.

use_bias? : boolean, whether the layer uses a bias vector.
Default value “True”.
Β paddingΒ :Β boolean, False = β€œvalid” means no padding. True = β€œsame” results in padding with zeros evenly to the left/right or up/down of the input such that output has the same height/width dimension as the input.
Default value β€œFalse”.
Β data_format :Β enum, one ofΒ channels_lastΒ orΒ channels_firstΒ (default) . The ordering of the dimensions in the inputs.Β channel_lastΒ corresponds to inputs with shapeΒ (batch, steps, features)Β whileΒ channels_firstΒ corresponds to inputs with shapeΒ (batch, features, steps).
Default value β€œchannels_first”.
depthwise_filter_initializerΒ : enum, for the depthwise convolution kernel.
Default value “glorot_uniform”.
pointwise_filter_initializerΒ : enum, for the pointwise convolution kernel.
Default value “glorot_uniform”.
Β bias_initializerΒ :Β enum, initializer for the bias vector.
Default value β€œzero”.
Β training?Β :Β boolean, whether the layer is in training mode (can store data for backward).
Default value β€œTrue”.
Β store?Β :Β boolean, whether the layer stores the last iteration gradient (accessible via the β€œget_gradients” function).
Default value β€œFalse”.
Β update?Β :Β boolean, whether the layer’s variables should be updated during backward. Equivalent to freeze the layer.
Default value β€œTrue”.
Β lda_coeff :Β float, defines the coefficient by which the loss derivative will be multiplied before being sent to the previous layer (since during the backward run we go backwards).
Default value β€œ1”.

 

Output parameters

 

SeparableConv2D out : layer separable conv 2D architecture.

Example

All these exemples are snippets PNG, you can drop these Snippet onto the block diagram and get the depicted code added to your VI (Do not forget to install HAIBAL library to run it).

SeparableConv2D layer inside TimeDistributed layer

1 – Generate a set of data

We generate an array of data of type single and shape [batch_size = 10, time = 6, channels = 7, rows = 5, cols = 3].

2 – Define graph

First, we define the first layer of the graph which is an Input layer (explicit input layer method). This layer is setup as an input array shaped [time = 6, channels = 7, rows = 5, cols = 3].
Then, we add to the graph the TimeDistributed layer which we setup with a SeparableConv2D layer using the define method.

3 – Run graph

We call the forward method and retrieve the result with the β€œPrediction 5D” method.
This method returns two variables, the first one is the layer information (cluster composed of the layer name, the graph index and the shape of the output layer) and the second one is the prediction with a shape of [batch_size, time, filters, new_rows, new_cols].

 

Tags:
Table of Contents