galilai-group/lejepa

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1322 stars · 115 forks observed · 2026-08-28

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

LeJEPA is a Python framework for scalable, theoretically grounded self-supervised representation learning based on Joint-Embedding Predictive Architectures (JEPAs). It introduces SIGReg (Sketched Isotropic Gaussian Regularization) to train vision models like ViTs and ConvNeXts without heuristic tricks.

Use cases

• pretrain vision models with self-supervised learning
• train a JEPA on ImageNet without heuristics
• learn image embeddings for few-shot classification
• evaluate pretrained representations on downstream datasets
• reproduce self-supervised learning research benchmarks

When to choose

• you want provable, heuristic-free self-supervised pretraining for vision models
• you need strong image representations with fewer pretraining epochs
• you are researching JEPA-style representation learning

When to avoid

• you need supervised classification out of the box
• you lack GPU resources for large ViT/ConvNeXt training
• you need NLP or multimodal self-supervised learning

Facets

library · maturity active

machine-learning deep-learning image-processing machine-learning deep-learning computer-vision artificial-intelligence python self-supervised-learning jeepa representation-learning vision-transformer sigreg research-code gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="galilai-group/lejepa")