lucidrains/byol-pytorch

Health v2 · maintenance only

Adoption not part of the score

1900 stars · 248 forks observed · 2026-10-21

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A PyTorch library implementing the Bootstrap Your Own Latent (BYOL) self-supervised learning method from DeepMind. It wraps any image-based neural network so it can be trained on unlabeled images without contrastive learning or negative pairs.

Use cases

• train image encoders without labeled data
• pretrain a resnet on unlabeled images
• self-supervised representation learning in pytorch
• improve downstream supervised performance with unlabeled images
• avoid paying for image labels
• implement BYOL without contrastive negative pairs

When to choose

• you have lots of unlabeled images and few labels
• you want a simple, pip-installable BYOL implementation to wrap an existing PyTorch network
• you want to avoid contrastive learning setups like SimCLR

When to avoid

• your data is not images
• you need a full training pipeline with data loading and experiment tracking
• you need contrastive methods like SimCLR or MoCo specifically

Facets

library · maturity active

machine-learning deep-learning deep-learning machine-learning artificial-intelligence image-processing python self-supervised-learning byol pytorch representation-learning unsupervised-learning computer-vision

Member repositories

For agents

markdown · JSON · MCP: product_card(name="lucidrains/byol-pytorch")