LibreYOLO

English | 简体中文

⭐ Support LibreYOLO. The best way to help is to star the repo. Feel free to open an issue if you encounter problems or have suggestions, and code contributions are very welcome (see CONTRIBUTING.md). If you want to contribute economically, that is very welcome: see the sponsorship program.

An MIT-licensed computer vision library. Detection, segmentation, pose, depth, OCR and a dozen more tasks behind one small API, with training and export included rather than sold separately. Reads common YOLO-format datasets, so existing workflows port over with minimal changes.

Install

pip install libreyolo

from libreyolo import LibreYOLO, SAMPLE_IMAGE

model = LibreYOLO("LibreYOLO9t.pt")
result = model(SAMPLE_IMAGE, save=True)

git clone https://github.com/LibreYOLO/libreyolo.git
cd libreyolo
pip install -e .

One API, twenty tasks

The same three lines run every task. Only the checkpoint changes.

from libreyolo import LibreYOLO

LibreYOLO("LibreYOLO9t.pt")("street.jpg", save=True)             # detection
LibreYOLO("LibreDeepLabv3mv3-sem.pt")("street.jpg", save=True)   # semantic segmentation
LibreYOLO("LibreHRNetw32-pose.pt")("street.jpg", save=True)      # pose
LibreYOLO("LibreMiDaSs-depth.pt")("street.jpg", save=True)       # depth
LibreYOLO("LibreFeyNobgl-matte.pt")("portrait.jpg", save=True)   # background removal
LibreYOLO("LibreRTDETRv2n-obb.pt")("aerial.jpg", save=True)      # oriented boxes

Sources are not just files. Point it at a webcam, an RTSP stream, a video, a directory, a YouTube URL or your screen:

libreyolo predict --model yolo9-t --source 0 --show          # webcam
libreyolo predict --model yolo9-t --source rtsp://camera/1   # network camera
libreyolo predict --model yolo9-t --source screen            # screen capture

What ships

Per-family sizes, checkpoints and parity evidence live in the model reference.

Train

from libreyolo import LibreYOLO

model = LibreYOLO("LibreYOLO9t.pt")
model.train(data="dataset.yaml", epochs=100, imgsz=640)

libreyolo train --model yolo9-t --data dataset.yaml --epochs 100

Multi-GPU, LoRA, layer freezing, distillation, from-scratch training, and TensorBoard, MLflow, Weights & Biases, Comet, ClearML, Neptune and DVCLive logging are all supported. See the training guide.

Export and deploy

Twelve formats: ONNX, TorchScript, TensorRT, OpenVINO, CoreML, Core AI, TFLite (LiteRT), NCNN, MNN, RKNN, Paddle and ExecuTorch. Plus NVIDIA Triton serving and DeepStream config generation.

libreyolo export --model yolo9-t --format onnx

Support varies by family and task, see the export matrix.

Documentation

• Docs covers install, tasks, models, training, prediction, export and the CLI
• Benchmarks for independent numbers
• CHANGELOG.md for what changed

Sponsors

The code is written in free time. Training new models, such as LibreYOLO26 and MIT retrains of families with restrictive weights, costs rented GPU time, and the LibreYOLO Sponsorship Program pays for it. Companies get their logo here from $100 a month, and from $500 a month are named in every release and in the model card of every weight released while they sponsor. Hardware manufacturers can sponsor by sending a device.

No sponsors yet.

License

• Code: MIT License.
• Weights: pre-trained weights may inherit licensing from their original source, and not all of them are permissive. Check the license on the specific Hugging Face repo before you use one commercially. Every LibreYOLO Hugging Face model states its license.