UniVS<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Video

UniVS: Unified and Universal Video Segmentation.

Inheritance: NeuralNetworkBase<T> → SegmentationModelBase<T> → VideoSegmentationBase<T> → UniVS<T>

For Beginners

Unified video segmentation across tasks. Video instance and panoptic segmentation.

Common use cases:

  • Unified video segmentation across tasks
  • Video instance and panoptic segmentation
  • Referring video object segmentation
  • Multi-task video understanding

How It Works

Technical Details:

  • Prompts-as-queries unifying VIS, VPS, and VOS
  • Target-specific prompt generation
  • Per-object prompt memory: predicted masks are fed back as visual prompts
  • Handles both automatic and prompted segmentation

Reference: Li et al., “UniVS: Unified and Universal Video Segmentation with Prompts as Queries”, CVPR 2024.

Example

// Create a UniVS model for unified multi-task video segmentation
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.ImageClassification,
    inputHeight: 480, inputWidth: 480, inputDepth: 3, outputSize: 80);
var model = new UniVS<double>(architecture);
// Every tunable is now set through the options object; these are the defaults:
var custom = new UniVS<double>(architecture,
    options: new UniVSOptions { NumClasses = 80, DropRate = 0.1, ModelSize = UniVSModelSize.R50 });
            
// Or load a pre-trained ONNX model for video instance/panoptic segmentation
var onnxModel = new UniVS<double>(architecture, "univs.onnx");

Constructors

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

Methods

Method Summary
ModelMetadata<T> GetModelMetadata() Collects metadata describing this model’s configuration.
void AddCorrection(int objectId, Tensor<T> correctionMask)