lightly-ai/lightly

Health v2 · maintenance only

Adoption not part of the score

3797 stars · 355 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

LightlySSL is a Python library built on PyTorch for self-supervised learning on images, offering modular implementations of methods like SimCLR, MoCo, DINO, and masked autoencoders. It includes a CLI for training models and generating embeddings without labels, plus a model zoo of pretrained checkpoints.

Use cases

• train a vision model without labels
• generate image embeddings from unlabeled photos
• pretrain a backbone for object detection
• run self-supervised learning on satellite imagery
• curate and understand a large unlabeled image dataset
• train SimCLR or MoCo on CIFAR-10

When to choose

• you have large unlabeled image datasets and want label-free pretraining
• you want a modular PyTorch framework for SSL research
• you need image embeddings for data curation or active learning

When to avoid

• you need a one-command pretraining product with Docker support (use LightlyTrain)
• your task is NLP or tabular data rather than images
• you want a fully managed data labeling platform (use LightlyStudio)

Facets

library · maturity active

machine-learning deep-learning image-processing cli computer-vision deep-learning machine-learning image-processing python self-supervised-learning contrastive-learning pytorch embeddings simclr moco dino model-zoo linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="lightly-ai/lightly")