VideoLLaMA3<T>

Models & Types · AiDotNet.VisionLanguage.VideoLanguage

VideoLLaMA 3: frontier multimodal for image and video.

Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → VideoLLaMA3<T>

Implements: IVideoLanguageModel<T>

For Beginners

VideoLLaMA 3 is Alibaba’s frontier multimodal model for advanced image and video understanding. Default values follow the original paper settings.

How It Works

VideoLLaMA 3 (Alibaba, 2025) is a frontier multimodal foundation model for both image and video understanding. It builds on the VideoLLaMA series with improved visual encoders, enhanced temporal modeling, and expanded training on large-scale image-video-text datasets for state-of-the-art performance on video comprehension benchmarks.

References:

  • Paper: “VideoLLaMA 3: Frontier Multimodal Foundation Models” (Alibaba, 2025)

Example

// Create a VideoLLaMA 3 model for frontier image and video understanding
// with advanced multimodal foundation model capabilities
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 VideoLLaMA3<double>(architecture, "videollama3.onnx");
            
// Training mode with native layers
var trainModel = new VideoLLaMA3<double>(architecture, new VideoLLaMA3Options());

Constructors

Constructor Summary
VideoLLaMA3(NeuralNetworkArchitecture<T> architecture, VideoLLaMA3Options options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
VideoLLaMA3(NeuralNetworkArchitecture<T> architecture, string modelPath, VideoLLaMA3Options 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 VideoLLaMA 3’s any-resolution visual tokenizer with adaptive spatial token merging.
Tensor<T> GenerateFromVideo(IReadOnlyList<Tensor<T>> frames, string prompt) Generates output from video frames using VideoLLaMA 3’s adaptive spatial-temporal token merging.