Type Alias: SemanticSegmenter<L>

SemanticSegmenter<L> = object

Defined in: extensions/cv/tasks/semanticSegmentation.ts:78

Semantic segmentation task runner.

Type Parameters

L

L extends PropertyKey = string

The type representing the segmentation labels.

Properties

dispose()

readonly dispose: () => void

Defined in: extensions/cv/tasks/semanticSegmentation.ts:82

Releases all allocated native resources.

Returns

void

---

segment()

readonly segment: (input, colormap?) => Promise<SemanticSegmentationResult<L>>

Defined in: extensions/cv/tasks/semanticSegmentation.ts:107

Runs semantic segmentation asynchronously.

For models returning logits for all classes, this performs an argmax over the class dimensions to identify the highest-probability class per pixel, then maps each class index to an RGBA color from the colormap. The final mapped colormap is returned in the result.

For models returning only a single logit value for the positive class, this applies a sigmoid activation, normalizes the probability values to grayscale (0-255), and converts the output to a grayscale RGBA mask. No colormap is applied, and the colormap in the result is undefined.

Parameters

input

ImageBuffer

The input image buffer to segment.

colormap?

Partial<ColorMap<L>>

Optional partial color mapping overrides for labels (applicable only for multi-class models). If not provided, a default colormap is automatically generated with distinct, high-contrast colors (with the first label defaulting to transparent). If a partial map is provided, any labels omitted from it will default to being rendered as fully transparent.

Returns

Promise<SemanticSegmentationResult<L>>

A promise resolving to the segmentation result.

Throws

With code RESOURCE_BUSY if the model is in use, or RESOURCE_DISPOSED if disposed.

---

segmentWorklet()

readonly segmentWorklet: (input, colormap?) => SemanticSegmentationResult<L>

Defined in: extensions/cv/tasks/semanticSegmentation.ts:116

Synchronous version of segment to be executed directly on the caller or worklet thread.

Parameters

input

ImageBuffer

colormap?

Partial<ColorMap<L>>

Returns

SemanticSegmentationResult<L>