ContrastiveEncoderOptions

Options & Configuration · AiDotNet.VisionLanguage.Encoders

Base configuration options for contrastive vision-language encoders (CLIP-family models).

Inheritance: ModelOptions → ContrastiveEncoderOptions

Known derived types: ALIGNOptions, BASICOptions, BiomedCLIPOptions, CLIPAOptions, DFNCLIPOptions, DeCLIPOptions, EVACLIPOptions, FLIPOptions, LLM2CLIPOptions, LiTOptions, MedCLIPOptions, MetaCLIPOptions, OpenCLIPOptions, RegionCLIPOptions, RemoteCLIPOptions, SigLIP2Options, SigLIPOptions

For Beginners

These settings control how the model processes images and text. The most important ones are:

  • ImageSize: The size images are resized to before processing (e.g., 224x224 pixels)
  • EmbeddingDim: How many numbers represent each image or text (bigger = more detail)
  • Temperature: How “sharp” the similarity comparisons are (lower = more decisive)

How It Works

Contrastive encoders learn a shared embedding space for images and text via contrastive learning. They share common hyperparameters: image/text embedding dimensions, projection dimension, temperature, and image preprocessing settings.

Constructors

Constructor Summary
ContrastiveEncoderOptions() Initializes a new instance with default values.
ContrastiveEncoderOptions(ContrastiveEncoderOptions other) Initializes a new instance by copying from another instance.

Properties

Property Summary
OnnxModelOptions OnnxOptions { get; set; } Gets or sets ONNX runtime options.
TextEncoderVariant TextEncoderVariant { get; set; } Gets or sets the text encoder variant.
ViTVariant VisionEncoderVariant { get; set; } Gets or sets the vision encoder variant.
double DropoutRate { get; set; } Gets or sets the dropout rate for regularization.
double LabelSmoothing { get; set; } Gets or sets the label smoothing factor.
double LearningRate { get; set; } Gets or sets the initial learning rate.
double Temperature { get; set; } Gets or sets the temperature parameter for contrastive loss.
double WeightDecay { get; set; } Gets or sets the weight decay for regularization.
double[] ImageMean { get; set; } Gets or sets the per-channel mean for image normalization.
double[] ImageStd { get; set; } Gets or sets the per-channel standard deviation for image normalization.
int ImageSize { get; set; } Gets or sets the input image size (height = width).
int MaxSequenceLength { get; set; } Gets or sets the maximum text token sequence length.
int NumTextHeads { get; set; } Gets or sets the number of text attention heads.
int NumTextLayers { get; set; } Gets or sets the number of text transformer layers.
int NumVisionHeads { get; set; } Gets or sets the number of vision attention heads.
int NumVisionLayers { get; set; } Gets or sets the number of vision transformer layers.
int PatchSize { get; set; } Gets or sets the ViT patch size in pixels.
int ProjectionDim { get; set; } Gets or sets the shared projection dimension for the joint embedding space.
int TextEmbeddingDim { get; set; } Gets or sets the text encoder embedding dimension.
int VisionEmbeddingDim { get; set; } Gets or sets the vision encoder embedding dimension.
int VisionFfnMultiplier { get; set; } Gets or sets the feed-forward hidden dimension multiplier for vision encoder.
int VocabSize { get; set; } Gets or sets the vocabulary size.
int WarmUpSteps { get; set; } Gets or sets the warm-up steps for the learning rate scheduler.
string ImageEncoderModelPath { get; set; } Gets or sets the path to the image encoder ONNX model.
string TextEncoderModelPath { get; set; } Gets or sets the path to the text encoder ONNX model.