TransMix<T>

Models & Types · AiDotNet.Augmentation.Image

TransMix (Chen et al., 2022) - attention-guided mixing for Vision Transformers. Creates patch-level mixing masks based on attention-like scores.

Inheritance: AugmentationBase<T, ImageTensor<T>> → LabelMixingAugmentationBase<T, ImageTensor<T>> → ImageMixingAugmenterBase<T> → TransMix<T>

Constructors

Constructor Summary
TransMix(double alpha, int patchSize, double probability)

Properties

Property Summary
int PatchSize { get; }

Methods

Method Summary
IDictionary<string, object> GetParameters()
ImageTensor<T> ApplyTransMix(ImageTensor<T> image1, ImageTensor<T> image2, Vector<T> labels1, Vector<T> labels2, AugmentationContext<T> context) Mixes two images using patch-level attention-guided masks.