MetaCLIP<T>

Models & Types · AiDotNet.VisionLanguage.Encoders

MetaCLIP (Demystifying CLIP Data) reproduces and improves CLIP via a fully transparent data curation pipeline.

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

Implements: IContrastiveVisionLanguageModel<T>

How It Works

MetaCLIP (Xu et al., 2023) provides a fully transparent CLIP data curation algorithm reproducing CLIP’s results from scratch and outperforming OpenAI CLIP at scale.

References:

  • Paper: “Demystifying CLIP Data” (Xu et al., 2023)

Constructors

Constructor Summary
MetaCLIP(NeuralNetworkArchitecture<T> architecture, MetaCLIPOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
MetaCLIP(NeuralNetworkArchitecture<T> architecture, string imageEncoderModelPath, MetaCLIPOptions options)

Properties

Property Summary
T Temperature { get; }
int EmbeddingDimension { get; }
int MaxSequenceLength { get; }
int ProjectionDimension { get; }
int TextEmbeddingDimension { get; }

Methods

Method Summary
Dictionary<string, T> ZeroShotClassify(Tensor<T> image, string[] labels)
ModelMetadata<T> GetModelMetadata()
T ComputeSimilarity(Tensor<T> image, string text)
Tensor<T> EncodeImage(Tensor<T> image)
Tensor<T> EncodeText(string text)
Tensor<T>[] EncodeTexts(string[] texts)