Intel® Geti™ is an end-to-end platform for building computer vision AI models, targeting engineers and domain experts who need to accelerate model development with less data. It offers an interactive workflow for rapid iteration, smart annotation tools, and supports multiple computer vision tasks, ultimately producing production-ready models optimized for Intel's OpenVINO™ toolkit.

How It Works

Geti employs a microservice and event-driven architecture orchestrated by Kubernetes. Key features include active learning for efficient data selection, AI-assisted smart annotations (including visual prompting with Meta AI's Segment Anything Model) to reduce manual effort, and task chaining for modular model development. This approach aims to significantly decrease the time and data required for state-of-the-art computer vision model creation.

Quick Start & Requirements

• Installation via Intel® Geti™ Installer or Helm Charts.
• Requires Kubernetes deployment. Access via web browser.
• Supports Python 3.10+, PyTorch 2.5+, and OpenVINO 2025.1.0.
• Refer to User Guide for detailed setup.

Highlighted Details

• Enables model building with as few as 10-20 images using active learning.
• Supports object detection, classification, segmentation, and anomaly detection tasks.
• Outputs optimized models for OpenVINO™ toolkit (Intel CPUs, GPUs, VPUs) or PyTorch format.
• Integrates with ecosystem tools like Datumaro for dataset management and Anomalib for anomaly detection.

Maintenance & Community

• Community support and bug reporting via GitHub Issues and Discussions.
• Contribution guide available for developers.

Licensing & Compatibility

• Intel® Geti™ platform licensed under LIMITED EDGE SOFTWARE DISTRIBUTION LICENSE.
• Fine-tuned models licensed under Apache License Version 2.0.
• Includes FFmpeg, licensed under LGPL and GPL; users are responsible for FFmpeg licensing compliance.

Limitations & Caveats

The platform's core licensing is restrictive for broader redistribution or commercial use outside of Intel's ecosystem. FFmpeg's dual licensing (LGPL/GPL) may also impose copyleft requirements depending on usage.