OMGLLaVA<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Referring

OMG-LLaVA: Bridging Image-Level, Object-Level, and Pixel-Level understanding.

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

For Beginners

Unified visual perception and understanding. Object-level and pixel-level reasoning.

Common use cases:

  • Unified visual perception and understanding
  • Object-level and pixel-level reasoning
  • Visual question answering with segmentation
  • Multi-granularity visual understanding

How It Works

Technical Details:

  • Universal segmentation model (Mask2Former) + LLM (LLaVA)
  • Bridges perception (segmentation) and understanding (reasoning)
  • Object-level queries from segmentation feed into LLM
  • Pixel-level grounding from LLM responses

Reference: Zhang et al., “OMG-LLaVA: Bridging Image-Level, Object-Level, Pixel-Level Reasoning and Understanding”, arXiv 2024.

Example

// Create an OMG-LLaVA model for multi-granularity visual understanding
// 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 OMGLLaVA<double>(architecture, textEncoder);
// Every tunable is now set through the options object; these are the defaults:
var custom = new OMGLLaVA<double>(architecture, textEncoder,
    options: new OMGLLaVAOptions { NumClasses = 1, DropRate = 0 });
            
// Or load a pre-trained ONNX model for visual question answering with segmentation
var onnxModel = new OMGLLaVA<double>(architecture, "omgllava.onnx", textEncoder, options: null);

Constructors

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

Properties

Property Summary
bool SupportsConversation { get; }

Methods

Method Summary
List<ReferringSegmentationResult<T>> SegmentVideoFromExpression(Tensor<T> frames, string expression)
ModelMetadata<T> GetModelMetadata() Collects metadata describing this model’s configuration.
ReferringSegmentationResult<T> SegmentFromConversation(Tensor<T> image, IReadOnlyList<ValueTuple<string, string>> conversationHistory, string currentQuery)