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).

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)

Optional extras

The base install covers YOLOv9 and the other core detectors, training, and inference. Add an extra when you need a heavier family or an export backend. Comma-separate to combine, for example pip install "libreyolo[rfdetr,onnx]".

Group	Extras
Export	onnx, tensorrt, openvino, coreml, coreai, tflite (alias litert), ncnn, mnn, paddle, executorch
Serving	triton
Models	rfdetr, vlm, sam, openvocab, clip, siglip2, eomt, midas, modus, sensenova, gaze
Training	lora, plots, tensorboard, mlflow, wandb, comet, clearml, neptune, dvclive
Speed	fast-eval, hub-kernels
Sources	stream
Everything	pip install "libreyolo[all]"

executorch, coreai and neptune are deliberately left out of all: they pin torch or protobuf in ways that would drag the rest of the environment with them. Full list and per-backend notes in the install guide.

Install from source

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

A plain clone checks out release, the stable branch matching the published package. For unreleased work, git checkout dev.

One API, seventeen 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

Task	Models
Detection	YOLOv9, RF-DETR, YOLOX, YOLO-NAS, D-FINE, DEIM, RT-DETR v1/v2/v4, RTMDet, PicoDet, YOLOv7, EfficientDet, and the classics: DETR, Deformable DETR, DINO-DETR, LW-DETR, Faster R-CNN, RetinaNet, SSD, FCOS, CenterNet
Tiny objects	Dome-DETR (aerial, drone, remote sensing)
Instance segmentation	RF-DETR, RTMDet, D-FINE, Mask R-CNN
Promptable segmentation	SAM, SAM 2, SAM 3, MobileSAM, EdgeTAM, PicoSAM3
Semantic segmentation	SegFormer, PIDNet, DeepLabv3, FCN, LingBot-Vision, DINOv2, EoMT
Panoptic segmentation	EoMT
Pose	RF-DETR, YOLO-NAS, HRNet, EC
Oriented boxes	RF-DETR, RT-DETRv2
Classification	MobileNetV4, ConvNeXt, EfficientNetV2, ResNet, ViT, Swin, DeiT, VGG, AlexNet, CLIP, SigLIP2, DINOv2
Depth	Depth Anything 3, Depth Anything V2, ZipDepth, MiDaS
Surface normals	MoGe-2
Edges	DexiNed, TEED
Embeddings	LibreFaceEmbedder, CLIP, SigLIP2, DINOv2
Body mesh	SAM 3D Body
Restoration	NAFNet, Real-ESRGAN, SwinIR
Background removal	BiRefNet, FeyNobg
OCR	PP-OCR
Point detection	FOMO, LocateAnything
Gaze	L2CS
Open vocabulary and VLMs	Grounding DINO, OWLv2, OmDet-Turbo, OV-DEIM, Florence-2, Kosmos-2, Qwen3-VL, InternVL3, LFM2-VL, SmolVLM2, MODUS

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

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.