ai.djl.modality.cv.util

Class NDImageUtils

java.lang.Object

ai.djl.modality.cv.util.NDImageUtils

public final class NDImageUtils
extends java.lang.Object

NDImageUtils is an image processing utility to load, reshape, and convert images using NDArray images.

Method Summary

Modifier and Type	Method and Description
static NDArray	centerCrop(NDArray image) Crops an image to a square of size min(width, height).
static NDArray	centerCrop(NDArray image, int width, int height) Crops an image to a given width and height from the center of the image.
static NDArray	crop(NDArray image, int x, int y, int width, int height) Crops an image with a given location and size.
static NDArray	normalize(NDArray input, float[] mean, float[] std) Normalizes an image NDArray of shape CHW or NCHW with mean and standard deviation.
static NDArray	normalize(NDArray input, float mean, float std) Normalizes an image NDArray of shape CHW or NCHW with a single mean and standard deviation to apply to all channels.
static NDArray	randomBrightness(NDArray image, float brightness) Randomly jitters image brightness with a factor chosen from [max(0, 1 - brightness), 1 + brightness].
static NDArray	randomColorJitter(NDArray image, float brightness, float contrast, float saturation, float hue) Randomly jitters the brightness, contrast, saturation, and hue of an image.
static NDArray	randomFlipLeftRight(NDArray image) Randomly flip the input image left to right with a probability of 0.5.
static NDArray	randomFlipTopBottom(NDArray image) Randomly flip the input image top to bottom with a probability of 0.5.
static NDArray	randomHue(NDArray image, float hue) Randomly jitters image hue with a factor chosen from [max(0, 1 - hue), 1 + hue].
static NDArray	randomResizedCrop(NDArray image, int width, int height, double minAreaScale, double maxAreaScale, double minAspectRatio, double maxAspectRatio) Crop the input image with random scale and aspect ratio.
static NDArray	resize(NDArray image, int size) Resizes an image to the given width and height.
static NDArray	resize(NDArray image, int width, int height) Resizes an image to the given width and height.
static NDArray	resize(NDArray image, int width, int height, Image.Interpolation interpolation) Resizes an image to the given width and height with given interpolation.
static NDArray	toTensor(NDArray image) Converts an image NDArray from preprocessing format to Neural Network format.

Methods inherited from class java.lang.Object

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

Method Detail

resize

public static NDArray resize(NDArray image,
                             int size)

Resizes an image to the given width and height.

Parameters:

image - the image to resize

size - the desired size

Returns:

the resized NDList

resize

public static NDArray resize(NDArray image,
                             int width,
                             int height)

Resizes an image to the given width and height.

Parameters:

image - the image to resize

width - the desired width

height - the desired height

Returns:

the resized NDList

resize

public static NDArray resize(NDArray image,
                             int width,
                             int height,
                             Image.Interpolation interpolation)

Resizes an image to the given width and height with given interpolation.

Parameters:

image - the image to resize

width - the desired width

height - the desired height

interpolation - the desired interpolation

Returns:

the resized NDList

normalize

public static NDArray normalize(NDArray input,
                                float mean,
                                float std)

Normalizes an image NDArray of shape CHW or NCHW with a single mean and standard deviation to apply to all channels.

Parameters:

input - the image to normalize

mean - the mean to normalize with (for all channels)

std - the standard deviation to normalize with (for all channels)

Returns:

the normalized NDArray

See Also:

normalize(NDArray, float[], float[])

normalize

public static NDArray normalize(NDArray input,
                                float[] mean,
                                float[] std)

Normalizes an image NDArray of shape CHW or NCHW with mean and standard deviation.

Given mean (m1, ..., mn) and standard deviation (s1, ..., sn for n channels, this transform normalizes each channel of the input tensor with: output[i] = (input[i] - m1) / (s1).

Parameters:

input - the image to normalize

mean - the mean to normalize with for each channel

std - the standard deviation to normalize with for each channel

Returns:

the normalized NDArray

toTensor

public static NDArray toTensor(NDArray image)

Converts an image NDArray from preprocessing format to Neural Network format.

Converts an image NDArray of shape HWC in the range [0, 255] to a DataType.FLOAT32 tensor NDArray of shape CHW in the range [0, 1].

Parameters:

image - the image to convert

Returns:

the converted image

centerCrop

public static NDArray centerCrop(NDArray image)

Crops an image to a square of size min(width, height).

Parameters:

image - the image to crop

Returns:

the cropped image

See Also:

centerCrop(NDArray, int, int)

centerCrop

public static NDArray centerCrop(NDArray image,
                                 int width,
                                 int height)

Crops an image to a given width and height from the center of the image.

Parameters:

image - the image to crop

width - the desired width of the cropped image

height - the desired height of the cropped image

Returns:

the cropped image

crop

public static NDArray crop(NDArray image,
                           int x,
                           int y,
                           int width,
                           int height)

Crops an image with a given location and size.

Parameters:

image - the image to crop

x - the x coordinate of the top-left corner of the crop

y - the y coordinate of the top-left corner of the crop

width - the width of the cropped image

height - the height of the cropped image

Returns:

the cropped image

randomFlipLeftRight

public static NDArray randomFlipLeftRight(NDArray image)

Randomly flip the input image left to right with a probability of 0.5.

Parameters:

image - the image with HWC format

Returns:

the flipped image

randomFlipTopBottom

public static NDArray randomFlipTopBottom(NDArray image)

Randomly flip the input image top to bottom with a probability of 0.5.

Parameters:

image - the image with HWC format

Returns:

the flipped image

randomResizedCrop

public static NDArray randomResizedCrop(NDArray image,
                                        int width,
                                        int height,
                                        double minAreaScale,
                                        double maxAreaScale,
                                        double minAspectRatio,
                                        double maxAspectRatio)

Crop the input image with random scale and aspect ratio.

Parameters:

image - the image with HWC format

width - the output width of the image

height - the output height of the image

minAreaScale - minimum targetArea/srcArea value

maxAreaScale - maximum targetArea/srcArea value

minAspectRatio - minimum aspect ratio

maxAspectRatio - maximum aspect ratio

Returns:

the cropped image

randomBrightness

public static NDArray randomBrightness(NDArray image,
                                       float brightness)

Randomly jitters image brightness with a factor chosen from [max(0, 1 - brightness), 1 + brightness].

Parameters:

image - the image with HWC format

brightness - the brightness factor from 0 to 1

Returns:

the transformed image

randomHue

public static NDArray randomHue(NDArray image,
                                float hue)

Randomly jitters image hue with a factor chosen from [max(0, 1 - hue), 1 + hue].

Parameters:

image - the image with HWC format

hue - the hue factor from 0 to 1

Returns:

the transformed image

randomColorJitter

public static NDArray randomColorJitter(NDArray image,
                                        float brightness,
                                        float contrast,
                                        float saturation,
                                        float hue)

Randomly jitters the brightness, contrast, saturation, and hue of an image.

Parameters:

image - the image with HWC format

brightness - the brightness factor from 0 to 1

contrast - the contrast factor from 0 to 1

saturation - the saturation factor from 0 to 1

hue - the hue factor from 0 to 1

Returns:

the transformed image