public final class SegmentationMask

Object
SegmentationMask

Dense per-pixel foreground confidence mask. The confidence array is row-major with exactly width * height values in the range 0..1. It is defensively copied on construction and access.

The mask’s resolution is the segmentation model’s, not the source image’s, so it is normally smaller than the frame it describes. cutOut(Image, float) and toMaskImage(int) rescale it for you; when reading getConfidence() directly, index it with getWidth() and getHeight() rather than the frame size.

SelfieSegmenter segmenter = new SelfieSegmenter();
segmenter.process(VisionImage.encoded(jpeg)).ready(mask -> {
    Image person = mask.cutOut(EncodedImage.create(jpeg), 0.6f);
    background.getStyle().setBgImage(person);
}).except(error -> Log.e(error));

Constructors

`public SegmentationMask(int width, int height, float[] confidence)`	Creates a row-major confidence mask without backend metadata.
`public SegmentationMask(int width, int height, float[] confidence,` `VisionMetadata metadata)`	Creates a row-major confidence mask with backend diagnostics.

Methods

`public int getWidth()`
`public int getHeight()`
`public float[] getConfidence()`
`public float getConfidenceAt(int x, int y)`	Reads one mask pixel.
`public` `Image cutOut(Image source, float threshold)`	Keeps the foreground of an image and makes the rest transparent.
`public` `Image toMaskImage(int rgb)`	Renders the mask itself as a translucent tint, for drawing over the frame it describes as a foreground highlight.
`public` `VisionMetadata getMetadata()`

Inherited methods

From Object

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

Constructor details

SegmentationMask

public SegmentationMask(int width, int height, float[] confidence)

Creates a row-major confidence mask without backend metadata.

Parameters

width int: mask width in pixels
height int: mask height in pixels
confidence float[]: one foreground probability per pixel, defensively copied

Throws

IllegalArgumentException: if either dimension is negative, the pixel count exceeds the maximum Java array length, or confidence does not contain exactly one value per pixel

SegmentationMask

public SegmentationMask(int width, int height, float[] confidence, VisionMetadata metadata)

Creates a row-major confidence mask with backend diagnostics.

Parameters

width int: mask width in pixels
height int: mask height in pixels
confidence float[]: one foreground probability per pixel, defensively copied
metadata VisionMetadata: backend details, or null

Throws

IllegalArgumentException: if either dimension is negative, the pixel count exceeds the maximum Java array length, or confidence does not contain exactly one value per pixel

Method details

getWidth

public int getWidth()

Returns

mask width, which may differ from source image width

getHeight

public int getHeight()

Returns

mask height, which may differ from source image height

getConfidence

public float[] getConfidence()

Returns

defensive copy of row-major foreground confidences

getConfidenceAt

public float getConfidenceAt(int x, int y)

Reads one mask pixel.

Parameters

x int: column in the range 0 to getWidth() - 1
y int: row in the range 0 to getHeight() - 1

Returns

foreground probability in the range 0..1

Throws

IndexOutOfBoundsException: if the position is outside the mask

cutOut

public Image cutOut(Image source, float threshold)

Keeps the foreground of an image and makes the rest transparent.

The mask is sampled with nearest-neighbour scaling, so the result has source’s dimensions no matter what resolution the segmentation model produced. Pixels whose confidence is below threshold become fully transparent; the rest keep their original color and alpha.

Parameters

source Image: the image the mask was computed from
threshold float: minimum foreground confidence to keep, clamped to 0..1

Returns

a new image with the background removed

Throws

NullPointerException: if source is null
IllegalArgumentException: if this mask has no pixels

toMaskImage

public Image toMaskImage(int rgb)

Renders the mask itself as a translucent tint, for drawing over the frame it describes as a foreground highlight.

Each pixel takes rgb as its color and the mask confidence as its alpha, so a fully confident foreground pixel is opaque and a background pixel is invisible. The result has the mask’s own resolution; draw it scaled to the frame’s bounds.

Parameters

rgb int: tint color as 0xRRGGBB; any alpha in the value is ignored in favor of the mask

Returns

a new image at this mask’s resolution

Throws

IllegalArgumentException: if this mask has no pixels

getMetadata

public VisionMetadata getMetadata()

Returns

backend metadata, or null when manually constructed