Self Supervised Learning

64 public types in this namespace, organized by kind.

Models & Types (41)

Type Summary
BYOLLoss<T> BYOL (Bootstrap Your Own Latent) Loss - a simple cosine similarity loss.
BYOL<T> BYOL: Bootstrap Your Own Latent - Self-supervised learning without negative samples.
BarlowTwinsLoss<T> Barlow Twins Loss - redundancy reduction loss for self-supervised learning.
BarlowTwins<T> Barlow Twins: Self-Supervised Learning via Redundancy Reduction.
BenchmarkResult<T> Result from a single benchmark evaluation.
BenchmarkSuite<T> Complete benchmark suite results.
CenteringMechanism<T> Centering mechanism for preventing collapse in self-distillation methods.
DINOLoss<T> DINO (Self-Distillation with No Labels) Loss for self-supervised learning.
DINO<T> DINO: Self-Distillation with No Labels - a self-supervised method for Vision Transformers.
FineTuningResult<T> Result from fine-tuning an SSL pretrained encoder.
InfoNCELoss<T> InfoNCE (Noise Contrastive Estimation) Loss for contrastive learning.
KNNEvaluator<T> k-Nearest Neighbors (k-NN) evaluation for SSL representation quality.
LinearEvalResult<T> Results from linear evaluation.
LinearEvaluator<T> Linear evaluation protocol for assessing SSL representation quality.
LinearProjector<T> Linear projection head for self-supervised learning.
MAEReconstructionLoss<T> MAE (Masked Autoencoder) Reconstruction Loss for self-supervised learning.
MAE<T> MAE: Masked Autoencoder for Self-Supervised Vision Learning.
MLPProjector<T> Multi-layer perceptron (MLP) projection head for self-supervised learning.
MemoryBank<T> FIFO memory queue for storing embeddings in contrastive learning.
MoCoV2<T> MoCo v2: Improved Baselines with Momentum Contrastive Learning.
MoCoV3<T> MoCo v3: An Empirical Study of Training Self-Supervised Vision Transformers.
MoCo<T> MoCo: Momentum Contrast for Unsupervised Visual Representation Learning.
MomentumEncoder<T> Momentum-updated encoder for self-supervised learning methods.
NTXentLoss<T> Normalized Temperature-scaled Cross Entropy Loss (NT-Xent) for contrastive learning.
SelfSupervisedLearningAugmentationContext<T> Context for SSL augmentation operations.
SelfSupervisedLearningAugmentationPolicies<T> Provides standard augmentation policies for self-supervised learning methods.
SelfSupervisedLearningFineTuningPipeline<T> Pipeline for fine-tuning SSL pretrained encoders on downstream tasks.
SelfSupervisedLearningMetricReport<T> Complete SSL metrics report.
SelfSupervisedLearningMetrics<T> Metrics for monitoring and evaluating self-supervised learning.
SelfSupervisedLearningPretrainingPipeline<T> High-level pipeline for SSL pretraining.
SelfSupervisedLearningPretrainingResult<T> The outcome of a self-supervised pretraining pass: the loss trajectory, a representation-collapse check, and an optional linear-probe quality estimate.
SelfSupervisedLearningResult<T> Result from SSL pretraining containing the trained encoder and metrics.
SelfSupervisedLearningSession<T> Manages a self-supervised learning training session.
SelfSupervisedLearningStepResult<T> Result of a single SSL training step.
SelfSupervisedLearningTrainingHistory<T> Training history from SSL pretraining.
SimCLR<T> SimCLR: A Simple Framework for Contrastive Learning of Visual Representations.
SimSiam<T> SimSiam: Exploring Simple Siamese Representation Learning.
SymmetricProjector<T> Symmetric Projector Head for BYOL and SimSiam-style methods.
TemperatureScheduler Schedules temperature parameters during self-supervised learning training.
TransferBenchmark<T> Transfer learning benchmark for evaluating SSL representations on downstream tasks.
iBOT<T> iBOT: Image BERT Pre-Training with Online Tokenizer - combining DINO with masked image modeling.

Base Classes (3)

Type Summary
ContrastiveLossBase<T> Base class for the self-supervised and contrastive objectives.
SelfSupervisedLearningMethodBase<T> Abstract base class for self-supervised learning methods.
TeacherStudentSSL<T> Base class for teacher-student self-supervised learning methods.

Interfaces (5)

Type Summary
IDetachedTensor<T> Marker interface for tensors that should not receive gradients.
IMemoryBank<T> Defines the contract for memory banks used in contrastive learning methods.
IMomentumEncoder<T> Defines the contract for momentum-updated encoders used in SSL methods.
IProjectorHead<T> Defines the contract for projection heads used in self-supervised learning.
ISelfSupervisedLearningMethod<T> Defines the contract for self-supervised learning methods.

Enums (5)

Type Summary
BenchmarkProtocol Supported evaluation protocols for transfer learning benchmarks.
FineTuningStrategy Fine-tuning strategies for SSL pretrained encoders.
SelfSupervisedLearningCommunicationBackend Communication backends for distributed SSL training.
SelfSupervisedLearningOptimizerType Optimizer types optimized for SSL training.
TemperatureScheduleType Types of temperature scheduling strategies.

Structs (1)

Type Summary
DetachedTensor<T> A wrapper that marks a tensor as detached from the computation graph.

Options & Configuration (8)

Type Summary
BYOLConfig BYOL-specific configuration settings.
BarlowTwinsConfig Barlow Twins-specific configuration settings.
DINOConfig DINO-specific configuration settings.
FineTuningConfig Configuration for fine-tuning SSL pretrained encoders.
MAEConfig MAE-specific configuration settings.
MoCoConfig MoCo-specific configuration settings.
SelfSupervisedLearningConfig<T> Unified configuration for self-supervised learning with industry-standard defaults.
SelfSupervisedLearningDistributedConfig Configuration for distributed SSL training using DDP (Distributed Data Parallel).

Helpers & Utilities (1)

Type Summary
StopGradient<T> Provides stop-gradient operations for self-supervised learning.