Ringing in Release 2025.2 of Edge AI Libraries

[stevenhoenisch]

The 2025.2 release of Open Edge Platform and the Edge AI Libraries simplifies computer vision model fine tuning, speeds up video streams processing, and enhances provisioning for more streamlined and secure model deployment. In general, this release enables validation and support for Intel® Core™ Ultra Series 2 processors, utilizing built-in GPU and NPU to accelerate vision AI model training and fine-tuning. OpenVINO™ simplifies NPU deployment and enables secure Gen AI model deployment. And Geti™ introduces support for model fine-tuning on Intel® Core™ Ultra Processor Series 2.

This release also brings in broad camera support for multi-camera media streams and faster frames per second, and new sample applications using transformer-based models for speech recognition and real-time scene intelligence.

🎁 Here are some highlights of the release:

• Geti™: Enables model fine tuning on Intel® Core™ Ultra Series 2 processors with built-in GPU, putting more power into the hands of developers and enabling more continuous model fine tuning capabilities at the edge. Geti also adds a new state-of-the-art DEIM-DETR object detection model algorithm, which is especially helpful for small object detection use cases. Support for annotation shapes with holes enables model training for complex segmentation tasks for manufacturing and medical use cases. And further flexibility on data augmentation enables developers to fine-tune models by configuring selective data augmentation parameters.

• SceneScape: The scene intelligence microservice adds vertical scaling to track up to a hundred objects through time chunked processing. It also added a new object clustering service to analyze object properties as well a new mapping service to create 3d meshes and automatically calibrate cameras from camera inputs. These services simplify developer workflows for calibration as well as enabling scaling many object tracking to build a more comprehensive 3d scene intelligence application.

• OpenVINO™ toolkit: OpenVINO adds in more new model support, including a preview of Mixture-of-Experts (MoE) models, such as Qwen 3, optimized for CPUs and GPUs, as well as an encrypted blob format for secure deployment of OpenVINO GenAI models to help ensure that IP thefts for model weights and artifacts can be avoided in deployment.

• DL Streamer: Smart video processing, auto-optimizing camera settings, and unique AI-powered fingerprinting of objects for re-identification enables developers to enhance their video analytics application development workflows.

📣 Get the full story on all the enhancements in the release notes.

About Open Edge Platform

Open Edge Platform from Intel delivers a modular, composable stack of open-source software optimized to rapidly deploy, secure, and scale AI solutions at distributed edge sites.

Open Edge Platform comprises Edge Microvisor Toolkit, Edge Manageability Framework, Edge AI Libraries, and Edge AI Suites.

Vendors, developers, and technology partners can take part in the GitHub community for these solutions in various ways: contributing code, proposing a design, downloading and trying out releases, opening an issue, benchmarking application performance, and -- perhaps most important of all for building community -- participating in discussions. For more information on Open Edge Platform, check out these resources:

• Read and respond to blog posts
• Watch demo videos on YouTube

Participate in or Contribute to Edge AI Libraries

Here's how to participate in or contribute to Edge AI Libraries:

• Participate in discussions
• Open an issue
• Submit a pull request
• Read the Contribution Guide
• Report a security vulnerability
• Read and respond to blog posts
• View the documentation website
• Read the release notes

🏗️ To download a library, see the weekly build reports.