org.deeplearning4j.zoo.model.helper

Class FaceNetHelper

java.lang.Object

org.deeplearning4j.zoo.model.helper.FaceNetHelper

public class FaceNetHelper
extends Object

Inception is based on GoogleLeNet configuration of convolutional layers for optimization of resources and learning. You can use this module to add Inception to your own custom models. The GoogleLeNet paper: https://arxiv.org/abs/1409.4842 This module is based on the Inception GraphBuilderModule built for Torch and a Scala implementation of GoogleLeNet. https://github.com/Element-Research/dpnn/blob/master/Inception.lua https://gist.github.com/antikantian/f77e91f924614348ea8f64731437930d

Author:

Justin Long (crockpotveggies)

Constructor Summary

Constructors

Constructor and Description
FaceNetHelper()

Method Summary

Modifier and Type	Method and Description
static ComputationGraphConfiguration.GraphBuilder	appendGraph(ComputationGraphConfiguration.GraphBuilder graph, String moduleLayerName, int inputSize, int[] kernelSize, int[] kernelStride, int[] outputSize, int[] reduceSize, SubsamplingLayer.PoolingType poolingType, org.nd4j.linalg.activations.Activation transferFunction, String inputLayer)
static ComputationGraphConfiguration.GraphBuilder	appendGraph(ComputationGraphConfiguration.GraphBuilder graph, String moduleLayerName, int inputSize, int[] kernelSize, int[] kernelStride, int[] outputSize, int[] reduceSize, SubsamplingLayer.PoolingType poolingType, int pNorm, org.nd4j.linalg.activations.Activation transferFunction, String inputLayer)
static ComputationGraphConfiguration.GraphBuilder	appendGraph(ComputationGraphConfiguration.GraphBuilder graph, String moduleLayerName, int inputSize, int[] kernelSize, int[] kernelStride, int[] outputSize, int[] reduceSize, SubsamplingLayer.PoolingType poolingType, int poolSize, int poolStride, org.nd4j.linalg.activations.Activation transferFunction, String inputLayer)
static ComputationGraphConfiguration.GraphBuilder	appendGraph(ComputationGraphConfiguration.GraphBuilder graph, String moduleLayerName, int inputSize, int[] kernelSize, int[] kernelStride, int[] outputSize, int[] reduceSize, SubsamplingLayer.PoolingType poolingType, int pNorm, int poolSize, int poolStride, org.nd4j.linalg.activations.Activation transferFunction, String inputLayer) Appends inception layer configurations a GraphBuilder object, based on the concept of Inception via the GoogleLeNet paper: https://arxiv.org/abs/1409.4842
static SubsamplingLayer	avgPool7x7(int stride)
static SubsamplingLayer	avgPoolNxN(int size, int stride)
static BatchNormalization	batchNorm(int in, int out)
static ConvolutionLayer	c3x3reduce(int in, int out, double bias)
static ConvolutionLayer	c5x5reduce(int in, int out, double bias)
static ConvolutionLayer	conv1x1(int in, int out, double bias)
static ConvolutionLayer	conv3x3(int in, int out, double bias)
static ConvolutionLayer	conv5x5(int in, int out, double bias)
static ConvolutionLayer	conv7x7(int in, int out, double bias)
static ConvolutionLayer	convNxN(int reduceSize, int outputSize, int kernelSize, int kernelStride, boolean padding)
static ConvolutionLayer	convNxNreduce(int inputSize, int reduceSize, int reduceStride)
static DenseLayer	fullyConnected(int in, int out, double dropOut)
static String	getModuleName()
static String	getModuleName(String layerName)
static SubsamplingLayer	maxPool3x3(int stride)
static SubsamplingLayer	maxPoolNxN(int size, int stride)
static SubsamplingLayer	pNormNxN(int pNorm, int size, int stride)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

FaceNetHelper

public FaceNetHelper()

Method Detail

getModuleName

public static String getModuleName()

getModuleName

public static String getModuleName(String layerName)

conv1x1

public static ConvolutionLayer conv1x1(int in,
                                       int out,
                                       double bias)

c3x3reduce

public static ConvolutionLayer c3x3reduce(int in,
                                          int out,
                                          double bias)

c5x5reduce

public static ConvolutionLayer c5x5reduce(int in,
                                          int out,
                                          double bias)

conv3x3

public static ConvolutionLayer conv3x3(int in,
                                       int out,
                                       double bias)

conv5x5

public static ConvolutionLayer conv5x5(int in,
                                       int out,
                                       double bias)

conv7x7

public static ConvolutionLayer conv7x7(int in,
                                       int out,
                                       double bias)

avgPool7x7

public static SubsamplingLayer avgPool7x7(int stride)

avgPoolNxN

public static SubsamplingLayer avgPoolNxN(int size,
                                          int stride)

pNormNxN

public static SubsamplingLayer pNormNxN(int pNorm,
                                        int size,
                                        int stride)

maxPool3x3

public static SubsamplingLayer maxPool3x3(int stride)

maxPoolNxN

public static SubsamplingLayer maxPoolNxN(int size,
                                          int stride)

fullyConnected

public static DenseLayer fullyConnected(int in,
                                        int out,
                                        double dropOut)

convNxN

public static ConvolutionLayer convNxN(int reduceSize,
                                       int outputSize,
                                       int kernelSize,
                                       int kernelStride,
                                       boolean padding)

convNxNreduce

public static ConvolutionLayer convNxNreduce(int inputSize,
                                             int reduceSize,
                                             int reduceStride)

batchNorm

public static BatchNormalization batchNorm(int in,
                                           int out)

appendGraph

public static ComputationGraphConfiguration.GraphBuilder appendGraph(ComputationGraphConfiguration.GraphBuilder graph,
                                                                     String moduleLayerName,
                                                                     int inputSize,
                                                                     int[] kernelSize,
                                                                     int[] kernelStride,
                                                                     int[] outputSize,
                                                                     int[] reduceSize,
                                                                     SubsamplingLayer.PoolingType poolingType,
                                                                     org.nd4j.linalg.activations.Activation transferFunction,
                                                                     String inputLayer)

appendGraph

public static ComputationGraphConfiguration.GraphBuilder appendGraph(ComputationGraphConfiguration.GraphBuilder graph,
                                                                     String moduleLayerName,
                                                                     int inputSize,
                                                                     int[] kernelSize,
                                                                     int[] kernelStride,
                                                                     int[] outputSize,
                                                                     int[] reduceSize,
                                                                     SubsamplingLayer.PoolingType poolingType,
                                                                     int pNorm,
                                                                     org.nd4j.linalg.activations.Activation transferFunction,
                                                                     String inputLayer)

appendGraph

public static ComputationGraphConfiguration.GraphBuilder appendGraph(ComputationGraphConfiguration.GraphBuilder graph,
                                                                     String moduleLayerName,
                                                                     int inputSize,
                                                                     int[] kernelSize,
                                                                     int[] kernelStride,
                                                                     int[] outputSize,
                                                                     int[] reduceSize,
                                                                     SubsamplingLayer.PoolingType poolingType,
                                                                     int poolSize,
                                                                     int poolStride,
                                                                     org.nd4j.linalg.activations.Activation transferFunction,
                                                                     String inputLayer)

appendGraph

public static ComputationGraphConfiguration.GraphBuilder appendGraph(ComputationGraphConfiguration.GraphBuilder graph,
                                                                     String moduleLayerName,
                                                                     int inputSize,
                                                                     int[] kernelSize,
                                                                     int[] kernelStride,
                                                                     int[] outputSize,
                                                                     int[] reduceSize,
                                                                     SubsamplingLayer.PoolingType poolingType,
                                                                     int pNorm,
                                                                     int poolSize,
                                                                     int poolStride,
                                                                     org.nd4j.linalg.activations.Activation transferFunction,
                                                                     String inputLayer)

Appends inception layer configurations a GraphBuilder object, based on the concept of Inception via the GoogleLeNet paper: https://arxiv.org/abs/1409.4842

Parameters:

graph - An existing computation graph GraphBuilder object.

moduleLayerName - The numerical order of inception (like 2, 2a, 3e, etc.)

inputSize -

kernelSize -

kernelStride -

outputSize -

reduceSize -

poolingType -

poolSize -

poolStride -

inputLayer -

Returns: