ThreeDGraphLLM<T>

Models & Types · AiDotNet.VisionLanguage.ThreeD

3DGraphLLM: 3D scene graph as LLM input for spatial reasoning.

Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → ThreeDGraphLLM<T>

Implements: IThreeDVisionLanguageModel<T>

For Beginners

3DGraphLLM is a vision-language model that uses scene graphs for structured 3D spatial reasoning. Default values follow the original paper settings.

How It Works

3DGraphLLM (CogAI, 2025) uses 3D scene graphs as structured input for large language models. It constructs semantic graphs from 3D scenes where nodes represent objects with their spatial properties and edges encode spatial relationships, providing the LLM with a structured representation for precise spatial reasoning, object counting, and relationship understanding.

References:

  • Paper: “3DGraphLLM: 3D Scene Graph as Input for Large Language Models (CogAI, 2025)”

Example

// Create a 3DGraphLLM model for 3D scene graph reasoning
// using semantic graphs as structured input for spatial understanding
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.TwoDimensional,
    taskType: NeuralNetworkTaskType.ImageClassification,
    inputHeight: 224, inputWidth: 224, inputDepth: 3, outputSize: 512);
            
// ONNX inference mode with pre-trained model
var model = new ThreeDGraphLLM<double>(architecture, "threedgraphllm.onnx");
            
// Training mode with native layers
var trainModel = new ThreeDGraphLLM<double>(architecture, new ThreeDGraphLLMOptions());

Constructors

Constructor Summary
ThreeDGraphLLM(NeuralNetworkArchitecture<T> architecture, ThreeDGraphLLMOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
ThreeDGraphLLM(NeuralNetworkArchitecture<T> architecture, string modelPath, ThreeDGraphLLMOptions options)

Properties

Property Summary
int DecoderEmbeddingDim { get; }
int EmbeddingDimension { get; }
int MaxGenerationLength { get; }
int MaxPoints { get; }
int PointChannels { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
Tensor<T> EncodeImage(Tensor<T> image)
Tensor<T> GenerateFrom3D(Tensor<T> pointCloud, string prompt) Processes 3D point cloud using 3DGraphLLM’s scene graph approach.
Tensor<T> GenerateFromImage(Tensor<T> image, string prompt) Generates from 2D image using 3DGraphLLM’s scene graph approach.