ai.konduit.serving.data.image.convert.config

Class ImageNormalization

java.lang.Object

ai.konduit.serving.data.image.convert.config.ImageNormalization

public class ImageNormalization
extends Object

• NONE: No image normalization will be applied
• SCALE: Divide images by maxValue/2, or divide by 255/2 if maxValue is not specified, then subtract 1.0, to give output in range [-1,1]. This is the default.
• SCALE_01: Divide images by maxValue, or divide by 255 if maxValue is not specified, to give output in range [0,1].
• SUBTRACT_MEAN: Subtract the channels by the provided meanRgb array, with values [meanRed, meanGreen, meanBlue]. out = (in - mean) for each channel. Note that if the output format is in BGR format, the meanRgb value should still be provided in RGB order
• STANDARDIZE: Subtract the channels by the provided meanRgb array, and then divide by stdRgb, where meanRgb is [meanRed, meanGreen, meanBlue], and stdRgb is [standardDeviationRed, standardDeviationGreen, standardDeviationBlue]. out = (in - mean)/std for each channel. Note that if the output format is in BGR format, the meanRgb and stdRgb values should still be provided in RGB order.
• INCEPTION: Applies inception preprocessing for inference/evaluation as described here: inception_preprocessing.py Specifically: preprocess_for_eval method: scale to [-1, 1] range. In practice this is done by dividing by 255 (assuming pixels are in range 0 to 255) to give [0, 1] then subtracting 0.5 and multiplying by 2 to give [-1, 1]. Note uses maxValue (like SCALE) if provided.
• VGG_SUBTRACT_MEAN: As per SUBTRACT_MEAN but the fixed values [meanRed, meanGreen, meanBlue] = [123.68, 116.779, 103.939]. Note the meanRgb array with these values need not be provided explicitly. If the output format is BGR, these are appropriately reordered before applying to the channels.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	ImageNormalization.Type

Field Summary

Fields

Modifier and Type	Field and Description
protected static double[]	IMAGE_NET_MEAN_RGB
protected static double[]	IMAGE_NET_STD_RGB
ImageNormalization.Type	type
protected static double[]	VGG_MEAN_RGB

Constructor Summary

Constructors

Constructor and Description
ImageNormalization()
ImageNormalization(ImageNormalization.Type type)

Method Summary

Modifier and Type	Method and Description
static double[]	getImagenetMeanRgb()
static double[]	getImageNetStdRgb()
static double[]	getVggMeanRgb()

Methods inherited from class java.lang.Object

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

Field Detail

VGG_MEAN_RGB

protected static final double[] VGG_MEAN_RGB

IMAGE_NET_MEAN_RGB

protected static final double[] IMAGE_NET_MEAN_RGB

IMAGE_NET_STD_RGB

protected static final double[] IMAGE_NET_STD_RGB

type

public ImageNormalization.Type type

Constructor Detail

ImageNormalization

public ImageNormalization()

ImageNormalization

public ImageNormalization(ImageNormalization.Type type)

Method Detail

getVggMeanRgb

public static double[] getVggMeanRgb()

getImagenetMeanRgb

public static double[] getImagenetMeanRgb()

getImageNetStdRgb

public static double[] getImageNetStdRgb()