#3535th of 53 in Training and fine-tuning
lightly-train
Python library for pretraining, distilling and fine-tuning computer vision models
- Stars
- 1.7k
- License
- AGPL-3.0
- Last commit
- Nov 2026
- Last release
- Oct 2026
- Language
- Python
1,395 on 21 Apr 20261,734 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
LightlyTrain is a Python package for training computer vision models: self-supervised pretraining and distillation (DINOv2/v3) on unlabeled data, plus fine-tuning for object detection, instance, semantic and panoptic segmentation, and depth estimation. Detection uses LTDETR/LTDETRv2 and segmentation uses EoMT, with ONNX and TensorRT export and C++ inference recipes. It is AGPL-3.0; commercial use needs a separate license.
Who it is for: ML engineers training detection and segmentation models for edge deployment
Strengths
- Covers pretraining, distillation and fine-tuning for detection and several segmentation tasks in one package
- Pretrained COCO checkpoints with published mAP and TensorRT T4 latency tables
- ONNX and TensorRT export, plus C++ inference recipes for ONNX Runtime and TensorRT
- Single pip install; runs on Windows, Linux and macOS with Python 3.8+
Weaknesses
- AGPL-3.0; production or proprietary use requires a commercial license from Lightly
- Vision only; no language model training or fine-tuning
- Benchmarks are on COCO, Cityscapes and NYUv2; custom-data results are not documented in the README
- Pretraining results depend on data; the README points struggling users to paid expert contact
What it needs
- GPU optional
- Needs PyTorch, TensorRT (optional, for export), ONNX Runtime (optional)
- Models: DINOv2, DINOv3, LTDETR, LTDETRv2, EoMT
Quick answers
Is lightly-train open source?
Yes. Its code is on GitHub under the AGPL-3.0 license.
Does lightly-train need a GPU?
No. It runs on a CPU and can use a GPU when you have one.
Also in Training and fine-tuning
See all 53| Rank | Project | Score |
|---|---|---|
| 1 | unslothDesktop and web app to run and fine-tune local models | 88 out of 100 |
| 2 | peftPython library for LoRA and other parameter-efficient fine-tuning of Transformers models | 88 out of 100 |
| 3 | deepspeedPyTorch library for distributed training of very large models with ZeRO | 86 out of 100 |
| #4 | verlReinforcement learning training library for LLM post-training with PPO and GRPO | 86 out of 100 |
| #5 | trlPost-training library with SFT, DPO, GRPO and KTO trainers | 86 out of 100 |