open-edge-platform/geti_v2

What it solves

Geti solves the complexity of building computer vision AI models by providing a seamless, end-to-end workflow. It addresses the challenges of manual data annotation, the need for large datasets, and the difficulty of deploying optimized models to edge hardware.

How it works

The platform manages the complete AI model lifecycle through a cloud-native, distributed architecture. It uses active learning to select the most impactful samples from a dataset, reducing the amount of manual data required. It also utilizes state-of-the-art models like the Segment Anything Model (SAM) for visual prompting and smart annotations. Once trained, models can be optimized for the Intel XPU portfolio using the OpenVINO toolkit.

Who it’s for

Geti is designed for developers and domain experts looking to build vision solutions for industrial and commercial use cases, such as detecting defective parts on production lines, automating inventory management, or enhancing robotics.

Highlights

• Smart Annotations: Supports bounding boxes, segmentation, and more with AI assistance.
• Interactive Training: Uses active learning to achieve high accuracy with as few as 10-20 images.
• Task Chaining: Allows users to decouple complex tasks into smaller, specialized models (e.g., detection followed by classification).
• Production-Ready: Outputs optimized models for OpenVINO or PyTorch for deployment on Intel CPUs, GPUs, and VPUs.