GPT4Point<T>
Models & Types · AiDotNet.VisionLanguage.ThreeD
GPT4Point: unified point-language understanding and generation.
Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → GPT4Point<T>
Implements: IThreeDVisionLanguageModel<T>
For Beginners
GPT4Point is a vision-language model for understanding and generating 3D point clouds using natural language. Default values follow the original paper settings.
How It Works
GPT4Point (2024) is a unified framework for point cloud-language understanding and generation. It bridges 3D point cloud representations with language models through a point cloud encoder that projects geometric features into the LLM’s embedding space, enabling both point cloud captioning and text-conditioned 3D point cloud generation in a single model.
References:
- Paper: “GPT4Point: A Unified Framework for Point-Language Understanding and Generation (Various, 2024)”
Example
// Create a GPT4Point model for unified point cloud-language understanding
// and 3D point cloud generation from text descriptions
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 GPT4Point<double>(architecture, "gpt4point.onnx");
// Training mode with native layers
var trainModel = new GPT4Point<double>(architecture, new GPT4PointOptions());
Constructors
| Constructor | Summary |
|---|---|
GPT4Point(NeuralNetworkArchitecture<T> architecture, GPT4PointOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) |
|
GPT4Point(NeuralNetworkArchitecture<T> architecture, string modelPath, GPT4PointOptions 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 GPT4Point’s Point-QFormer alignment approach. |
Tensor<T> GenerateFromImage(Tensor<T> image, string prompt) |
Generates from 2D image using GPT4Point’s Point-QFormer alignment. |