SceneLLM<T>

Models & Types · AiDotNet.VisionLanguage.ThreeD

Scene-LLM: voxel-based 3D scene understanding with language models.

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

Implements: IThreeDVisionLanguageModel<T>

For Beginners

Scene-LLM is a vision-language model for voxel-based 3D scene understanding with language-guided spatial reasoning. Default values follow the original paper settings.

How It Works

Scene-LLM (2024) extends language models for 3D visual understanding and reasoning using voxel-based scene representations. It discretizes 3D environments into voxel grids with semantic features, enabling language-guided spatial reasoning about object locations, room layouts, and navigation instructions in indoor environments.

References:

  • Paper: “Scene-LLM: Extending Language Model for 3D Visual Understanding and Reasoning (Various, 2024)”

Example

// Create a Scene-LLM model for voxel-based 3D scene understanding
// with language-guided spatial reasoning over indoor environments
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 SceneLLM<double>(architecture, "scenellm.onnx");
            
// Training mode with native layers
var trainModel = new SceneLLM<double>(architecture, new SceneLLMOptions());

Constructors

Constructor Summary
SceneLLM(NeuralNetworkArchitecture<T> architecture, SceneLLMOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
SceneLLM(NeuralNetworkArchitecture<T> architecture, string modelPath, SceneLLMOptions 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 SceneLLM’s hybrid scene representation.
Tensor<T> GenerateFromImage(Tensor<T> image, string prompt) Generates from 2D image using SceneLLM’s hybrid scene representation.