Qwen25VL<T>

Models & Types · AiDotNet.VisionLanguage.InstructionTuned

Qwen2.5-VL: visual agent with 1hr+ video, bounding box/point localization capabilities.

Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → Qwen25VL<T>

Implements: IInstructionTunedVLM<T>

For Beginners

Qwen2.5-VL extends Qwen2-VL with visual agent capabilities — it can not only understand images and videos but also interact with visual interfaces like a human user would. It supports understanding videos over 1 hour long, locating objects with bounding boxes and point coordinates, and excels at reading complex documents and charts. The model works as a visual agent that can navigate UIs, understand screenshots, and perform actions based on what it sees. Available in multiple sizes for different compute budgets. Default values follow the original paper settings.

How It Works

Qwen2.5-VL extends Qwen2-VL with enhanced visual agent capabilities including support for 1hr+ video understanding, bounding box and point localization, and stronger document/chart comprehension. Uses M-RoPE positional encoding and Qwen2.5 language backbone.

References:

  • Paper: “Qwen2.5-VL Technical Report” (2025)

Example

var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.TwoDimensional,
    taskType: NeuralNetworkTaskType.ImageClassification,
    inputHeight: 224, inputWidth: 224, inputDepth: 3, outputSize: 512);
var trainModel = new Qwen25VL<double>(architecture, new Qwen25VLOptions());

Constructors

Constructor Summary
Qwen25VL(NeuralNetworkArchitecture<T> architecture, Qwen25VLOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
Qwen25VL(NeuralNetworkArchitecture<T> architecture, string modelPath, Qwen25VLOptions options)

Properties

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

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
Tensor<T> Chat(Tensor<T> image, IEnumerable<ValueTuple<string, string>> conversationHistory, string userMessage)
Tensor<T> EncodeImage(Tensor<T> image)
Tensor<T> GenerateFromImage(Tensor<T> image, string prompt) Generates text using Qwen2.5-VL’s visual-agent architecture.