Duo deliver 20 times more AI performance for edge devices

The push towards widespread adoption of Artificial Intelligence (AI) in consumer devices continues at a relentless pace. However, cloud-based deep learning on battery-powered devices is plagued with issues, including latency, security and the need for a constant, reliable internet connection. Implementing the intelligence on the device itself – or on the edge – eliminates all of these issues. Highly efficient computer vision processors are necessary to meet the stringent power requirements and specialised deep learning software is crucial in delivering the accuracy and performance needed for cloud-based systems.

Targeting embedded devices, Brodmann17 has developed a specialised deep learning technology for visual recognition aimed at edge-based artificial intelligence. Using patent-pending techniques, Brodmann17’s deep learning architecture generates smaller neural-networks that are faster and more accurate than any other network generated on the market. Through the collaboration with Brodmann17, licensees of the CEVA-XM platforms and their customers will be able to use Brodmann17’s deep learning object detection that achieves state of the art accuracy on the CEVA-XM at a rate of 100 frames per second. This equates to 170% better performance than the same software running on the NVIDIA Jetson TX2 AI Supercomputer. Comparing to the popular combination today of Faster-RCNN algorithm over NVIDIA TX2 it is an improvement of 20 times (2000%) in frames per second.

“Our patent-pending deep learning vision software is a perfect fit for the many CEVA customers and OEMs using CEVA-XM platforms to add intelligence to their devices,” said Adi Pinhas, CEO of Brodmann17. “This first-of-its-kind combination of hardware and software achieves real time performance that supports multi-cameras with a single DSP or higher resolutions.”

“To truly maximise the performance and capabilities of AI, in mass-market devices, it requires not just application-specific hardware like our CEVA-XM platforms, but also neural networks that are trained to be run efficiently on the edge embedded devices,” said IlanYona, Vice President of the vision business unit at CEVA. “Brodmann17’s deep learning software provides the capability to create extremely light, accurate and flexible networks, trained from the ground up with embedded in mind. We’re delighted to partner with them and bring their unique capabilities to the CEVA-XM ecosystem.”

Brodmann17 will present and demonstrate their deep learning software together with the CEVA-XM4 imaging and vision platform at the CEVA Technology Symposium Series in China and Taiwan, 23rd to 27th October 2017. Visitors to the symposium can meet computer vision experts from CEVA and Brodmann17 at these events.