org.tribuo.interop.tensorflow.example

Class CNNExamples

java.lang.Object

org.tribuo.interop.tensorflow.example.CNNExamples

public abstract class CNNExamples
extends Object

Static factory methods which produce Convolutional Neural Network architectures.

Method Summary

Modifier and Type	Method and Description
static GraphDefTuple	buildLeNetGraph(String inputName, int imageSize, int pixelDepth, int numOutputs) Builds a LeNet 5 style CNN (usually used for MNIST).

Methods inherited from class java.lang.Object

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

Method Detail

buildLeNetGraph

public static GraphDefTuple buildLeNetGraph(String inputName,
                                            int imageSize,
                                            int pixelDepth,
                                            int numOutputs)

Builds a LeNet 5 style CNN (usually used for MNIST).

Expects there to only be a single colour pixel (i.e., a grayscale image). Operates on square images.

Unlike the original LeNet 5 it uses ReLU activations.

Parameters:

inputName - The input placeholder name.

imageSize - The image width and height.

pixelDepth - The maximum pixel value (usually 255).

numOutputs - The number of output dimensions (usually 10 for MNIST).

Returns:

A GraphRecord containing the LeNet graph and the relevant op names.