SelfSupervisedLearningAugmentationPolicies<T>

Models & Types · AiDotNet.SelfSupervisedLearning

Provides standard augmentation policies for self-supervised learning methods.

For Beginners

Self-supervised learning methods rely heavily on data augmentation to create different “views” of the same image. These views should be different enough to be challenging, but similar enough that they can still be recognized as the same image.

How It Works

Common augmentations for SSL:

Method-specific policies:

Constructors

Constructor Summary
SelfSupervisedLearningAugmentationPolicies(int? seed) Initializes a new instance of the SelfSupervisedLearningAugmentationPolicies class.

Methods

Method Summary
SelfSupervisedLearningAugmentationPolicies<T> ForBYOL(int? seed) Creates a standard BYOL augmentation policy.
SelfSupervisedLearningAugmentationPolicies<T> ForDINO(int? seed) Creates a standard DINO augmentation policy.
SelfSupervisedLearningAugmentationPolicies<T> ForSimCLR(int? seed) Creates a standard SimCLR augmentation policy.
Tensor<T> ApplyMinimal(Tensor<T> image) Applies minimal augmentation (used by MAE which relies on masking rather than augmentation).
ValueTuple<Tensor<T>, Tensor<T>> ApplyBYOL(Tensor<T> image) Applies BYOL-style asymmetric augmentation.
ValueTuple<Tensor<T>, Tensor<T>> ApplyMoCoV2(Tensor<T> image) Applies MoCo v2 style augmentation.
ValueTuple<Tensor<T>, Tensor<T>> ApplySimCLR(Tensor<T> image) Applies SimCLR-style augmentation to create two views.
ValueTuple<Tensor<T>[], Tensor<T>[]> ApplyDINO(Tensor<T> image, int numGlobalCrops, int numLocalCrops) Applies DINO-style multi-crop augmentation.