Edge Impulse taps Nvidia Omniverse for edge AI

Edge Impulse taps Nvidia Omniverse for edge AI

New Products |
By Nick Flaherty


Edge Impulse has released a new suite of tools developed on Nvidia’s Omniverse and AI platforms to bring the latest AI models to a wider range of edge microcontrollers.

This allows the Nvidia Tao models from GPUs to run on edge devices such as MPUs and MCUs, significantly speeding up the maturation of those models. This capability was included in Toolkit5.0 back in June 2023.

The Omniverse integration accelerates the use of large Nvidia GPU-trained models on affordable MCUs and MPUs with AI accelerators. Users now have access to a large library of production-tested pretrained models directly in the Edge Impulse platform, and Edge Impulse’s EON Tuner simplifies selection of the optimal model for each application.

This supports creating accurate, custom, production-ready computer vision models that can be seamlessly deployed to edge-optimized hardware, including the ARM Cortex-M based NXP I.MXRT1170, Alif E3, STMicro STM32H747AI, and Renesas CK-RA8D1. The Edge Impulse platform allows users to now provide their own custom data with GPU-trained Tao models like YOLO and RetinaNet, optimizing them for deployment.

This also enables deployment of large-scale Nvidia models to ARM-based devices, opening up a significant universe of hardware that can now be augmented with best-in-breed AI and ML models.

“The advent of generative AI and the growth of IoT deployments means the industry must evolve to run AI models at the edge,” said Paul Williamson, senior vice president and general manager, IoT Line of Business at ARM. “Nvidia and Edge Impulse have now made it possible to deploy state-of-the-art computer vision models on a broad range of technology based on Arm Cortex-M and Cortex-A CPUs and Arm Ethos-U NPUs, unlocking a multitude of new AI use cases at the edge.”

Edge Impulse has developed applications for synthetic data and testing environments for the edge with Nvidia Omniverse for industries operating in complex industrial, remote, or sensitive environments, where obtaining real-world data can be costly, time-consuming, create privacy concerns, or simply cannot account for all types of scenarios. 

The Omniverse Replicator framework for developing custom synthetic data generation pipelines can be integrated into existing workflows to generate highly realistic, physically based datasets tailored to train computer vision models. Now, with Omniverse Replicator combined with Edge Impulse, users can rapidly create professional-grade industrial ML models that can run on resource-constrained devices, for use cases such as visual inspection of manufacturing production lines to detect defects, equipment malfunctions, or surgery inventory object detection to prevent postoperative complications.

This allows customers to reduce physical prototyping and testing costs via virtual tools and speed up development time and experimentation. It can simulate sensor and model behaviour, and test MCU compatibility as well as use synthetic data to fortify model reliability and create difficult-to-replicate scenarios.

“Working closely with NVIDIA has enabled us to significantly expand the practical applications of AI on the edge for critical business use cases in industrial productivity, healthcare, and much more. For the first time, NVIDIA’s state-of-the-art machine learning research and model architectures can be deployed on any device under the sun, from the smallest microcontrollers to the latest GPUs and neural accelerators,” said Jan Jongboom, co-founder and CTO of Edge Impulse.

“Omniverse and Tao have incredibly simplified the creation of all computer vision models, including the latest generative AI models,” said Deepu Talla, vice president of robotics and edge computing at Nvidia. “Edge Impulse is integrating this powerful capability into easy-to-use workflows for the hundreds of billions of IoT and edge devices, including MCUs, accelerators and CPUs.”


If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles