OpenVocabSegmentationOutput<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Common

Output for open-vocabulary and referring segmentation models.

For Beginners

Open-vocabulary segmentation lets you segment objects described by arbitrary text. This output maps each text query to its corresponding mask(s) in the image. Referring segmentation adds reasoning capability — the model can also explain what it found.

Constructors

Constructor Summary
OpenVocabSegmentationOutput()

Properties

Property Summary
List<GroundingToken> GroundingTokens { get; set; } Grounding tokens connecting text spans to image regions (for grounded models like GLaMM).
Tensor<T> BoundingBoxes { get; set; } Per-query bounding boxes [numQueries, 4] as (x1, y1, x2, y2).
Tensor<T> Masks { get; set; } Per-query masks [numQueries, H, W].
Tensor<T> SemanticMap { get; set; } Combined semantic map [H, W] where each pixel is assigned to the best-matching query.
TimeSpan InferenceTime { get; set; } Inference time.
double[] Scores { get; set; } Per-query confidence scores.
double[] SimilarityScores { get; set; } Text-image similarity scores [numQueries] indicating how well each query matches the image.
string TextResponse { get; set; } Model’s textual response (for referring/reasoning segmentation models like LISA).
string[] Queries { get; set; } Text queries that were used.