Deep learning computer vision for Renesas MPU
Syntiant’s commercially available deep learning computer vision (CV) models can now be deployed on the integrated DRP-AI (Dramatically Reconfigurable Processor) core on the Renesas RZ/V2L microprocessor (MPU), allowing developers to add intelligence to camera-based devices quickly and without an energy overhead.
Designed for edge devices and optimized to reduce latency and memory footprint, Syntiant’s hardware-agnostic deep learning models can be used for multiple vision-based applications such as object detection, face recognition, pose estimation, background subtraction and image classification. Using Syntiant’s Inference Software Development Kit (SDK), the company’s CV algorithms are integrated onto a variety of hardware platforms and support both legacy and modern compute architectures, as well as solve critical problems directly on embedded devices to reduce cost and enable faster time to market.
“Working to deploy a solution that integrates our compute-efficient vision models on the Renesas RZ/V2L MPU enables next-gen embedded-vision applications for cameras used in the home, smart cities, retail POS systems and autonomous robots, among so many other use cases,” said Mallik Moturi, chief business officer at Syntiant. “Our combined technologies enable developers to quickly and easily create intelligent, energy-efficient devices that can operate in a wide range of environments.”
The Renesas RZ/V2L MPU is equipped with a Cortex®-A55 (1.2GHz) CPU and built-in DRP-AI accelerator that provides both real-time AI inference and image processing for camera support such as color correction and noise reduction. The platform also has a 16-bit DDR3L/DDR4 interface and a built-in 3D graphics engine with Arm Mali-G31 and video codec (H.264). Integrating Syntiant’s solutions on the DRP-AI core supports up to five different vision-based classes to be run in parallel, offering exceptional functionality and efficiency.
“Our ongoing collaboration with the team at Syntiant now includes integrating their computer vision models into our DRP-AI accelerator core for enhanced AI vision processing,” said Shigeki Kato, vice president of the Enterprise Infrastructure Business Division at Renesas. “This joint solution delivers both high performance and low power consumption and is able to reduce processing time by pre-processing images to reduce workload on the CPU. This is ideal for highly complex imaging processing such as barcode scanning and iris detection and extraction.”
With its unique, adaptive neural network structure, Syntiant’s CV models dynamically scale with the complexity of input. Models are further optimized to verify performance and reliability for high performance across numerous industries, ranging from smart home to retail analytics.