Upgrading edge deployments with powerful computer vision

Alif Semiconductor announced a significant breakthrough in implementing AI vision processing on edge MCUs. This is achieved by introducing full support for Nvidia's TAO model training toolkit on the Ensemble and Balletto MCU families and the Edge Impulse platform.

The TAO toolkit has generated significant enthusiasm among developers of edge AI devices due to its provision of extensive training datasets for common vision processing applications and its support for transfer learning from pre-trained models. This innovation promises to substantially lower the cost, time, and effort embedded device OEMs commonly invest in model development for AI vision applications.

Until now, deploying TAO-trained models on low-power MCUs for edge applications has been untested and unproven. Now, with the total integration of Nvidia TAO into its platform, Edge Impulse has established a streamlined process for deploying TAO models on the Alif Ensemble and Balletto families of MCUs and fusion processors. Fully integrated into the Edge Impulse platform, Alif's products feature the Arm Ethos-U55, an NPU for which Nvidia has optimised the TAO toolkit.

Embedded developers seeking to implement AI vision applications, such as people counting, intruder detection, or robotics, can now use the TAO training toolkit and its dataset, confidently deploying either a pre-trained TAO model or a custom model developed through transfer learning via Edge Impulse on Alif Ensemble or Balletto MCUs.

Henrik Flodell, senior marketing director at Alif Semiconductor, stated, "With Alif's Ai-optimised MCU ecosystem, high-end embedded vision processing is transitioning from large, expensive microprocessors to next-generation edge MCUs. The integration of the TAO toolkit by Edge Impulse has significantly streamlined the development and deployment of AI vision processing models on Alif MCUs."

Adam Benzion, SVP Partnerships at Edge Impulse, added, "The TAO toolkit accelerates the generation of effective ML models, but it doesn't address the deployment of these models on edge-optimised hardware. Together with Alif, we have solved this challenge by providing a fully integrated workflow from the TAO toolkit's pre-trained models to deployment on Alif edge MCUs."