BASIC<T>
Models & Types · AiDotNet.VisionLanguage.Encoders
BASIC (Combined Scaling for Zero-shot Transfer Learning) model using CoAtNet vision encoder.
Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → BASIC<T>
Implements: IContrastiveVisionLanguageModel<T>
For Beginners
BASIC pushes CLIP performance higher by combining three things at scale: a stronger vision architecture (CoAtNet, which mixes convolutions and attention), a larger dataset, and bigger batch sizes during training. The result is 85.7% zero-shot ImageNet accuracy. Default values follow the original paper settings.
How It Works
BASIC (Pham et al., 2022) combines CoAtNet (Convolution + Attention) vision encoder with large-scale contrastive image-text pre-training, achieving 85.7% zero-shot ImageNet via architecture+data+batch-size combined scaling.
References:
- Paper: “Combined Scaling for Zero-shot Transfer Learning” (Pham et al., 2022)
Example
var architecture = new NeuralNetworkArchitecture<double>(
inputType: InputType.TwoDimensional,
taskType: NeuralNetworkTaskType.ImageClassification,
inputHeight: 224, inputWidth: 224, inputDepth: 3, outputSize: 512);
var trainModel = new BASIC<double>(architecture, new BASICOptions());
Constructors
| Constructor | Summary |
|---|---|
BASIC(NeuralNetworkArchitecture<T> architecture, BASICOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) |
|
BASIC(NeuralNetworkArchitecture<T> architecture, string imageEncoderModelPath, BASICOptions 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) |