MAE<T>

Models & Types · AiDotNet.SelfSupervisedLearning

MAE: Masked Autoencoder for Self-Supervised Vision Learning.

Inheritance: ModelBase<T, Tensor<T>, Tensor<T>> → SelfSupervisedLearningMethodBase<T> → MAE<T>

For Beginners

MAE is a simple yet powerful self-supervised method. It randomly masks a large portion (75%) of image patches, encodes only the visible patches, and trains a decoder to reconstruct the original pixels of the masked patches.

How It Works

Key innovations:

Architecture:

Reference: He et al., “Masked Autoencoders Are Scalable Vision Learners” (CVPR 2022)

Best for: Efficient pretraining, generative understanding.

Pros: Efficient (only encode visible patches), scalable.

Cons: May require fine-tuning for best downstream performance.

Constructors

Constructor Summary
MAE(INeuralNetwork<T> encoder, INeuralNetwork<T> decoder, int patchSize, int imageSize, double maskRatio, SelfSupervisedLearningConfig<T> config) Initializes a new instance of the MAE class.

Properties

Property Summary
SelfSupervisedLearningMethodCategory Category { get; }
bool RequiresMemoryBank { get; }
bool UsesMomentumEncoder { get; }
double MaskRatio { get; } Gets the mask ratio (proportion of patches masked).
int PatchSize { get; } Gets the patch size.
string Name { get; }

Methods

Method Summary
MAE<T> Create(INeuralNetwork<T> encoder, INeuralNetwork<T> decoder, int patchSize, int imageSize, double maskRatio) Creates an MAE instance with default configuration.