IThreeDVisionLanguageModel<T>
Interfaces · AiDotNet.VisionLanguage.Interfaces
Interface for 3D vision-language models that understand point clouds, 3D scenes, and spatial relationships.
Implements: IGenerativeVisionLanguageModel<T>
Known implementations: GPT4Point<T>, LEOVL<T>, PointLLM<T>, SceneLLM<T>, ThreeDGraphLLM<T>, ThreeDLLM<T>
How It Works
3D vision-language models extend traditional 2D VLMs to process point clouds, voxel grids, and 3D scene representations. They enable spatial reasoning, 3D object grounding, and scene-level question answering.
Properties
| Property | Summary |
|---|---|
int MaxPoints { get; } |
Gets the maximum number of 3D points the model can process. |
int PointChannels { get; } |
Gets the number of channels per point (e.g., 3 for XYZ, 6 for XYZ+RGB). |
Methods
| Method | Summary |
|---|---|
Tensor<T> GenerateFrom3D(Tensor<T> pointCloud, string prompt) |
Processes a 3D point cloud and generates language output conditioned on a text prompt. |