geti

Overview of geti

What it is

Geti is an open-source, end-to-end platform for building and continuously improving computer vision models. It covers annotation, fine-tuning, optimization, deployment, and inference. It offers 1,600+ model variants for detection, classification, and segmentation, optimized by OpenVINO for Intel CPUs, GPUs, and NPUs. The platform includes a no-code application (Windows app or Docker) and a Python library called getitune for code-first workflows.

Key features

• 1,600+ model variants including RF-DETR, DINOv3 DETR, YOLO26, YOLOX, D-FINE, Mask R-CNN, and more, covering object detection, instance segmentation, and classification.
• Interactive end-to-end training: start with as few as 10-20 images and iterate through train-predict-annotate loops.
• Hardware-accelerated inference and model optimization: automatic export to OpenVINO format, support for Intel XPU, NVIDIA CUDA, and CPU; built-in accuracy-aware INT8 quantization.
• Integrated deployment and inference: build custom pipelines (source → model → sink) for live video streams, with exportable OpenVINO-optimized bundles.
• Multiple computer vision tasks: image classification, object detection, instance segmentation from the no-code interface, with more tasks available via the getitune library.
• Smart annotations: AI-assisted bounding box and polygon creation using Segment Anything Model, reducing manual annotation effort.
• Model and dataset management: track versions, link models to dataset revisions, view hyperparameters, and import/export in COCO, Pascal VOC, YOLO, and native formats.

Use cases

• Object detection and counting (e.g., counting items, defect localization, surveillance)
• Instance segmentation (e.g., medical imaging, robotics, area estimation)
• Image classification (e.g., defect classification, product categorization, content tagging)
• Building custom computer vision pipelines for edge deployment

Getting started

Install the getitune Python library: bash uv pip install getitune[xpu] --extra-index-url https://download.pytorch.org/whl/xpu # for Intel XPU uv pip install getitune[cuda] --extra-index-url https://download.pytorch.org/whl/cu128 # for NVIDIA CUDA uv pip install getitune # CPU-only Then train, export, and quantize a model: python from getitune.engine import create_engine from getitune.utils import list_models engine = create_engine(model=edgecrafter_m, data=./led_counting, work_dir=./my_workspace, device=auto) engine.train() exported_model_path = engine.export() openvino_engine = create_engine(model=exported_model_path, data=engine.datamodule) optimized_model_path = openvino_engine.optimize() predictions = openvino_engine.predict(checkpoint=optimized_model_path) For the no-code application, see the repository README for Windows MSIX, Docker, or script installation.

License and pricing

Licensed under Apache-2.0. Open source and free to use.

Based on the project's public README and repository metadata.