Shashi Chilappagari, Co-Founder and Chief Architect at DeGirum, demonstrates the company’s latest edge AI and vision technologies and products at the March 2023 Edge AI and Vision Innovation Forum. Specifically, Chilappagari demonstrates machine learning model multiplexing on the company’s ORCA AI hardware accelerator.
Developing real-world edge AI applications often requires using multiple machine learning (ML) models, running in series or in parallel. For example, an application to monitor vehicles may need to run a license plate detector followed by an optical character recognition (OCR) model. A gesture recognition application similarly might need to run a hand detector model followed by a gesture recognition model, for better accuracy. And a security application may need to run multiple models on the same video stream to gather different types of data.
Such applications greatly benefit from AI hardware accelerators and supporting software that can efficiently multiplex among the different ML models. This demo shows how DeGirum’s ORCA AI hardware accelerator efficiently multiplexes between different models. In the demo, three YOLOv5s-based ML models (person detection, face detection, and hand detection) run on two USB cameras, delivering 150 FPS of total performance.
