weka.dl4j.interpretability

Class ScoreCAM

java.lang.Object

weka.dl4j.interpretability.AbstractCNNSaliencyMapGenerator

weka.dl4j.interpretability.ScoreCAM

public class ScoreCAM
extends AbstractCNNSaliencyMapGenerator

Implementation of the ScoreCAM saliency map generation method.

Constructor Summary

Constructors

Constructor and Description
ScoreCAM()

Method Summary

Modifier and Type	Method and Description
java.awt.image.BufferedImage	generateHeatmapToImage(int[] targetClasses, java.lang.String[] classMap, boolean normalize) Generates heatmaps for the supplied classes, returning a human-viewable heatmap summary.
org.deeplearning4j.nn.conf.inputs.InputType.InputTypeConvolutional	getModelInputShape()
java.awt.image.BufferedImage	imageFromINDArray(org.nd4j.linalg.api.ndarray.INDArray array) Takes an INDArray containing an image loaded using the native image loader libraries associated with DL4J, and converts it into a BufferedImage.
void	processImage(java.io.File imageFile) Main processing entrypoint.
void	setModelInputShape(org.deeplearning4j.nn.conf.inputs.InputType.InputTypeConvolutional modelInputShape)

Methods inherited from class weka.dl4j.interpretability.AbstractCNNSaliencyMapGenerator

addIterationIncrementListener, addIterationsFinishedListeners, addIterationsStartedListener, getBatchSize, getComputationGraph, getImagePreProcessingScaler, getInputFilename, getModelName, isImageChannelsLast, setBatchSize, setComputationGraph, setImageChannelsLast, setImagePreProcessingScaler, setInputFilename, setModelName

Methods inherited from class java.lang.Object

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

Constructor Detail

ScoreCAM

public ScoreCAM()

Method Detail

processImage

public void processImage(java.io.File imageFile)

Description copied from class: AbstractCNNSaliencyMapGenerator

Main processing entrypoint. Takes an image as input, and performs the necessary processing on it. Note that this is performed separately to the actual saliency map generation. This method should only perform the expensive one-time processing on an image.

Specified by:

processImage in class AbstractCNNSaliencyMapGenerator

Parameters:

imageFile - File to process into saliency map.

generateHeatmapToImage

public java.awt.image.BufferedImage generateHeatmapToImage(int[] targetClasses,
                                                           java.lang.String[] classMap,
                                                           boolean normalize)

Description copied from class: AbstractCNNSaliencyMapGenerator

Generates heatmaps for the supplied classes, returning a human-viewable heatmap summary.

Specified by:

generateHeatmapToImage in class AbstractCNNSaliencyMapGenerator

Parameters:

targetClasses - Classes to generate saliency maps for.

classMap - Appropriate class map for the supplied classes.

normalize - Should we normalize the heatmap?

Returns:

Human-viewable heatmap summary.

getModelInputShape

public org.deeplearning4j.nn.conf.inputs.InputType.InputTypeConvolutional getModelInputShape()

Overrides:

getModelInputShape in class AbstractCNNSaliencyMapGenerator

setModelInputShape

public void setModelInputShape(org.deeplearning4j.nn.conf.inputs.InputType.InputTypeConvolutional modelInputShape)

Overrides:

setModelInputShape in class AbstractCNNSaliencyMapGenerator

imageFromINDArray

public java.awt.image.BufferedImage imageFromINDArray(org.nd4j.linalg.api.ndarray.INDArray array)

Takes an INDArray containing an image loaded using the native image loader libraries associated with DL4J, and converts it into a BufferedImage. The INDArray contains the color values split up across three channels (RGB) and in the integer range 0-255.

Parameters:

array - INDArray containing an image in order [N, C, H, W] or [C, H, W]

Returns:

BufferedImage