VideoChat2<T>

Models & Types · AiDotNet.VisionLanguage.VideoLanguage

VideoChat2: progressive video training with diverse data.

Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → VideoChat2<T>

Implements: IVideoLanguageModel<T>

For Beginners

VideoChat2 is a video-language model trained progressively on diverse video-text data for video conversation. Default values follow the original paper settings.

How It Works

VideoChat2 (Shanghai AI Lab, 2023) uses progressive video training with diverse multi-modal data. It trains through multiple stages starting from image-text alignment, then image instruction tuning, and finally video instruction tuning on diverse video datasets, building strong video conversation capabilities incrementally.

References:

  • Paper: “MVBench: A Comprehensive Multi-modal Video Understanding Benchmark” (Shanghai AI Lab, 2023)

Example

// Create a VideoChat2 model for video conversation
// with progressive training on diverse video-text data
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 VideoChat2<double>(architecture, "videochat2.onnx");
            
// Training mode with native layers
var trainModel = new VideoChat2<double>(architecture, new VideoChat2Options());

Constructors

Constructor Summary
VideoChat2(NeuralNetworkArchitecture<T> architecture, VideoChat2Options options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
VideoChat2(NeuralNetworkArchitecture<T> architecture, string modelPath, VideoChat2Options options)

Properties

Property Summary
int DecoderEmbeddingDim { get; }
int EmbeddingDimension { get; }
int MaxFrames { get; }
int MaxGenerationLength { get; }
string LanguageModelName { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
Tensor<T> EncodeImage(Tensor<T> image)
Tensor<T> GenerateFromImage(Tensor<T> image, string prompt) Generates from a single image using VideoChat2’s Q-Former visual-text cross-attention.
Tensor<T> GenerateFromVideo(IReadOnlyList<Tensor<T>> frames, string prompt) Generates output from video frames using VideoChat2’s Q-Former temporal aggregation.