Vision DSPs enable faster development of imaging applications

This enables faster development of computer vision and imaging applications on Tensilica Vision P-Series DSPs being deployed in applications processors for mobile, automotive, drone, security, Augmented Reality/Virtual Reality (AR/VR) and other markets.

In addition, Cadence released the OpenVX 1.1 Application Programming Kit (APK) for Tensilica Vision DSPs, which is available now.

“Cadence’s achievement of OpenVX 1.1 conformance for their Tensilica Vision P5 and P6 DSPs demonstrates their commitment to this open standard for vision processing and continues the momentum that has been building since OpenVX’s first release in 2014,” said Neil Trevett, President of the Khronos Group. “Khronos warmly welcomes Cadence’s ongoing contributions to the growth of OpenVX and their advancement of portable computer vision applications.”

Applications developed using the standard OpenVX 1.1 API can be compiled and run on Tensilica Vision DSPs without any code changes. Cadence’s OpenVX framework automatically schedules and executes the appropriate DMA transfers for efficient memory access, and runs highly optimised DSP vision-processing kernels in parallel with the DMA transfers. The Khronos OpenVX 1.1 compliance certification ensures that program results are consistent across various OpenVX 1.1 implementations.

“Cadence is the first company to pass OpenVX 1.1 compliance tests using a vision-optimised DSP, and releasing the OpenVX APK is a key milestone in making OpenVX and Tensilica’s Vision P-Series DSPs a defacto standard for computer vision, image processing and neural network applications,” said Steve Roddy, Senior Group Director, Tensilica marketing at Cadence. “Application developers can now reduce application software development time and easily take advantage of the massive parallel processing capability of our Tensilica Vision DSPs. OpenVX empowers application developers to redeploy applications across many different Tensilica processor-powered chips without recoding or re-optimising their applications.”

Vision-optimised DSPs are rapidly emerging in application-processor-class systems on chip, with three of the top five mobile application processor vendors now offering vision DSPs in their flagship devices. The OpenVX standard enables the deployment of algorithms on these complex, heterogeneous platforms featuring CPUs, GPUs and DSPs. Tensilica Vision P-Series DSPs offer class-leading performance of 256 MACs per cycle in less than 1mm² of silicon, delivering higher performance in a smaller area and with lower power than CPU- or GPU-based systems alone.

Developed by the nonprofit, member-funded consortium Khronos Group, OpenVX is an open, royalty-free standard for cross-platform acceleration of computer vision applications. OpenVX enables performance and power-optimised computer vision processing, especially important in embedded and real-time use cases such as face, body and gesture tracking, smart video surveillance, Advanced Driver Assistance Systems (ADAS), object and scene reconstruction, augmented reality, visual inspection, robotics and more. Frank Brill, Design Engineering Director at Cadence, is chairman of the OpenVX Working Group.

Cadence presenters will be discussing OpenVX computer vision library standards, as well as techniques to reduce power in embedded deep neural network applications and scalable neural network processors for embedded applications, at the Embedded Vision Alliance Summit being held 1^st-3^rd May 2017 at the Santa Clara Convention Center in Santa Clara, California.