DeCLIP<T>
Models & Types · AiDotNet.VisionLanguage.Encoders
DeCLIP (Data-efficient CLIP) model using self-supervised and nearest-neighbor supervision for data efficiency.
Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → DeCLIP<T>
Implements: IContrastiveVisionLanguageModel<T>
For Beginners
DeCLIP achieves CLIP-level performance with much less training data by extracting extra supervision from the data itself. It combines contrastive image-text matching with image self-supervision (SimSiam), text masked language modeling, and nearest- neighbor supervision — squeezing more learning from each image-text pair. Default values follow the original paper settings.
How It Works
DeCLIP (Li et al., ICLR 2022) achieves CLIP-level performance with less data by combining contrastive learning with image self-supervision (SimSiam), text MLM, and nearest-neighbor supervision.
References:
- Paper: “Supervision Exists Everywhere: A Data-Efficient Contrastive Language-Image Pre-training Paradigm” (Li et al., ICLR 2022)
Example
var architecture = new NeuralNetworkArchitecture<double>(
inputType: InputType.TwoDimensional,
taskType: NeuralNetworkTaskType.ImageClassification,
inputHeight: 224, inputWidth: 224, inputDepth: 3, outputSize: 512);
var trainModel = new DeCLIP<double>(architecture, new DeCLIPOptions());
Constructors
| Constructor | Summary |
|---|---|
DeCLIP(NeuralNetworkArchitecture<T> architecture, DeCLIPOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) |
|
DeCLIP(NeuralNetworkArchitecture<T> architecture, string imageEncoderModelPath, DeCLIPOptions 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) |