# lightly-ai/lightly-train

All-in-one training for vision models (YOLO, ViTs, RT-DETR, DINOv3): pretraining, fine-tuning, distillation.

Repository: https://github.com/lightly-ai/lightly-train
Canonical: https://ross.abutalabs.com/products/lightly-train
Homepage: https://docs.lightly.ai/train
Language: Python
License: AGPL-3.0
License Family: copyleft
Topics: computer-vision, contrastive-learning, deep-learning, distillation, embeddings, pretrained-models, python, pytorch, self-supervised, self-supervised-learning, dinov2, eomt, semantic-segmentation, vision-transformer, dinov3, rtdetrv2, yolo, object-detection, real-time, depth-estimation
Last push: 2026-08-26T08:38:59+00:00

## Health v2 (maintenance only)
Score: 85/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 95, longevity 36
- inputs: {"age_days": 510, "days_push": 7, "days_rel": 36, "gap_med": 13.5, "n_releases_24m": 31}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1652, forks 107 (observed 2026-08-28T04:05:17.026128+00:00)

## What it is
LightlyTrain is a Python framework for training computer vision models, covering pretraining of vision foundation models (DINOv2/v3) on unlabeled data, fine-tuning of transformer and YOLO models for detection and segmentation, and model distillation. It supports ONNX and TensorRT export for edge deployment and is licensed under AGPL-3.0 with a commercial license option.

## Use cases
- pretrain vision models on unlabeled images
- fine-tune YOLO for object detection
- train semantic segmentation models
- distill large vision models into smaller ones
- export detection models to ONNX and TensorRT
- train DINOv3 vision transformers on custom data
- train instance segmentation models for edge deployment
- compute image embeddings with self-supervised learning

## When to choose
- you need state-of-the-art pretraining or fine-tuning for vision models like YOLO, ViTs, or DETR variants
- you want to leverage unlabeled data via self-supervised pretraining
- you need distillation to shrink models for edge deployment
- you want built-in ONNX/TensorRT export for real-time inference

## When to avoid
- you need a permissive open-source license for proprietary commercial products without a commercial license
- you work outside computer vision (e.g., NLP or audio)
- you only need inference with existing pretrained models rather than training

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-training
- domain: computer-vision, machine-learning, deep-learning
- platform: python, cross-platform
- tags: self-supervised-learning, pretraining, fine-tuning, distillation, object-detection, semantic-segmentation, yolo, vision-transformer, dinov2, dinov3, pytorch, embeddings, contrastive-learning, onnx-export, tensorrt, gpu, docker

## Member repositories
- lightly-ai/lightly-train (main) score 85

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:17.026128+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:45:00.511267+00:00, confidence not recorded.
  - readme: https://github.com/lightly-ai/lightly-train (fetched 2026-08-28T04:05:17.026128+00:00, sha 73ecbebf47d6)
  - homepage: https://docs.lightly.ai/train (fetched 2026-08-29T11:18:09.966923+00:00, sha e7bb82bae82e)
- Data as of 2026-08-30T08:39:29.467469+00:00.