YOLOmatic is a production-focused Python CLI/TUI for configuring, training, fine-tuning, predicting, benchmarking, augmenting, converting, monitoring, and uploading computer-vision models. It covers YOLO26/12/11/10/9/8, YOLOX, RF-DETR, SAM 3.1, and Detectron2 from one terminal workflow.
- Interactive wizard UX: configure models, datasets, fine-tuning, prediction, benchmarking, augmentation, NDJSON conversion, TensorBoard, and upload flows without hand-writing boilerplate.
- Hardware-aware configs: CUDA, Apple Silicon MPS, CPU fallback, workers, batch sizes, and runtime preflight checks are handled before training starts.
- Safe offline augmentation: isolated worker processes create reproducible Albumentations datasets without replacing a prior output until the new one succeeds.
- 10+ model families in one CLI: Ultralytics YOLO, native RF-DETR, SAM 3.1, Detectron2, Roboflow upload, ClearML tracking, Labelbox conversion, and local benchmark reports.
uv tool install --python 3.12 yolomatic
yolomaticRepository development:
git clone https://github.com/shahabahreini/YOLOMatic.git
cd YOLOMatic
uv sync
uv run yolomaticCommon commands:
uv run yolomatic-train
uv run yolomatic-predict
uv run yolomatic-benchmark
uv run yolomatic-convert
uv run yolomatic-upload
uv run yolomatic-tensorboardUse the built-in semantic version command before creating a release tag:
uv run bump patch # or: minor / major
git tag vX.Y.Z
git push origin vX.Y.ZRelease tags must point to a commit already on main; GitHub Actions verifies
the tag and package version before publishing the wheel and source archive.
| Area | YOLOmatic support |
|---|---|
| Model families | YOLO26, YOLOv12, YOLO11, YOLOv10, YOLOv9, YOLOv8, YOLOX, RF-DETR, SAM 3.1, Detectron2 |
| Tasks | Detection, segmentation, classification, pose, OBB, open-vocabulary segmentation |
| Data | YOLO folders, COCO JSON, Labelbox NDJSON, Ultralytics-platform NDJSON |
| Training | Fresh training, checkpoint fine-tuning, resume, hardware preflight |
| Evaluation | mAP, F1, per-image rankings, UMAP vector analysis, interactive HTML report |
| Integrations | Roboflow, ClearML, TensorBoard, HuggingFace, Ultralytics |
| Capability | YOLOmatic | Ultralytics CLI | Roboflow | Hosted YOLO tools |
|---|---|---|---|---|
| Interactive terminal wizard | Yes | Limited | Web UI | Web UI |
| Multiple YOLO generations | Yes | Yes | Export/deploy focused | Varies |
| Native RF-DETR training | Yes | No | Deploy focused | Varies |
| SAM 3.1 workflows | Yes | No | Limited | Varies |
| Detectron2 training | Yes | No | No | Varies |
| Hardware-aware config generation | Yes | Manual | Abstracted | Abstracted |
| Labelbox/Ultralytics NDJSON conversion | Yes | No | Import focused | Varies |
| Local benchmark HTML reports | Yes | Partial metrics | Hosted analytics | Varies |
| Fully local workflow | Yes | Yes | No | No |
YOLOmatic is a Python 3.12 command-line toolkit with an interactive terminal UI for computer-vision training workflows.
YOLO26, YOLOv12, YOLO11, YOLOv10, YOLOv9, YOLOv8, YOLOX, RF-DETR, SAM 3.1, and Detectron2.
No. YOLOmatic also supports native RF-DETR training and fine-tuning, SAM 3.1 segmentation workflows, and Detectron2 training.
No. CUDA is recommended for training, but CPU and Apple Silicon MPS fallbacks are supported. YOLOmatic detects common CUDA/PyTorch mismatches and offers repair guidance.
Yes. It converts Labelbox and Ultralytics-platform NDJSON exports into YOLO or COCO datasets with concurrent image downloads. Ultralytics pose exports can be converted explicitly to YOLO Pose or COCO Pose while preserving keypoints.
Yes. uv run yolomatic-benchmark compares Ultralytics YOLO checkpoints and
exports such as .pt, ONNX, TensorRT .engine, TorchScript, and OpenVINO,
then verifies model/dataset task compatibility and evaluates train, valid, test,
or all groups. Detection and instance segmentation report mAP/F1, pose reports
box metrics, and semantic segmentation reports mIoU, pixel accuracy, and Dice.
Yes. uv run yolomatic-export exports trained YOLO .pt weights to TensorRT,
ONNX, OpenVINO, CoreML, TFLite, TorchScript, NCNN, RKNN, and other Ultralytics
formats. TensorRT workspace is user-controlled so it can be tuned for the GPU
that builds the engine. The export wizard reads the selected checkpoint details
first, then hides options that do not apply to that model task or export format.
Yes. YOLOmatic uploads YOLO checkpoints and deploys RF-DETR checkpoints through the upload TUI or optional post-training upload config.
| Configure Model | Smart Split |
|---|---|
 |
 |
- Full documentation
- Install guide
- Models guide
- YOLO guide
- Dataset guide
- Comparison
- Changelog
- Contributing
- Citation
@software{bahreini_jangjoo_yolomatic_2026,
author = {Bahreini Jangjoo, Shahab},
title = {{YOLOmatic}: automated computer-vision training CLI for YOLO, RF-DETR, SAM, and Detectron2},
year = {2026},
version = {4.4.0},
license = {Apache-2.0},
url = {https://github.com/shahabahreini/YOLOMatic},
note = {Python CLI/TUI for configuring, training, fine-tuning, predicting, benchmarking, and uploading computer-vision models}
}YOLOmatic is licensed under Apache 2.0.
YOLOmatic builds on the Python computer-vision ecosystem, including Ultralytics, RF-DETR, SAM/HuggingFace, Detectron2, PyTorch, Albumentations, ClearML, TensorBoard, Roboflow, and MkDocs Material.