VideoLISA<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Referring

Video-LISA: Language-instructed video segmentation with reasoning.

Inheritance: NeuralNetworkBase<T> → SegmentationModelBase<T> → ReferringSegmentationBase<T> → VideoLISA<T>

For Beginners

Language-instructed video segmentation. Temporal reasoning in video segmentation.

Common use cases:

  • Language-instructed video segmentation
  • Temporal reasoning in video segmentation
  • Referring video object segmentation
  • Conversational video understanding

How It Works

Technical Details:

  • Extends LISA to video with sparse token memory
  • Sparse Dense Sampling for temporal video understanding
  • One-Token-Seg-All: single token triggers multi-frame masks
  • Handles temporal reasoning queries in video

Reference: Bai et al., “One Token to Seg Them All: Language Instructed Reasoning Segmentation in Videos”, NeurIPS 2024.

Example

// Create a Video-LISA model for language-instructed video segmentation
// Any CLIP-style ITextEncoder supplies the text embeddings, e.g. a CLIP model from VisionLanguage.Encoders.
ITextEncoder<double> textEncoder = clipModel;
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.BinaryClassification,
    inputHeight: 1024, inputWidth: 1024, inputDepth: 3, outputSize: 1);
var model = new VideoLISA<double>(architecture, textEncoder);
// Every tunable is now set through the options object; these are the defaults:
var custom = new VideoLISA<double>(architecture, textEncoder,
    options: new VideoLISAOptions { NumClasses = 1, DropRate = 0 });
            
// Or load a pre-trained ONNX model for temporal reasoning segmentation
var onnxModel = new VideoLISA<double>(architecture, "videolisa.onnx", textEncoder, options: null);

Constructors

Constructor Summary
VideoLISA(NeuralNetworkArchitecture<T> architecture, ITextEncoder<T> textEncoder, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction, VideoLISAOptions options) Initializes VideoLISA in native (trainable) mode.
VideoLISA(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, ITextEncoder<T> textEncoder, VideoLISAOptions options) Initializes VideoLISA in ONNX (inference-only) mode.

Properties

Property Summary
bool SupportsConversation { get; }
bool SupportsVideoInput { get; }

Methods

Method Summary
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank)
ModelMetadata<T> GetModelMetadata() Collects metadata describing this model’s configuration.