Xilinx's reVISION Stack paves the way for a wide range of vision-oriented machine learning applications

     Xilinx, a global leader in All Programmable technologies and devices, introduces the Xilinx reVISION Stack, announcing the expansion of Xilinx technology into a wide range of vision-oriented machine learning applications. The introduction of the reVISION stack further complements and complements its recently released reconfigurable acceleration stack, significantly expanding the deployment of Xilinx technology in machine learning applications from end to cloud. The new reVISION stack supports a wider range of embedded software and systems engineers with little or no hardware design expertise, making it easier and faster to develop visually oriented intelligent systems using Silinx technology. Once the advantages of machine learning, computer vision, sensor fusion, and connectivity are combined, these engineers will reap the benefits.

     rReVISION is realizing a range of rapidly growing applications in a variety of markets, including the traditional high-end consumer market, automotive, industrial, medical, and aerospace and defense, as well as a new generation of applications such as collaborative robots, "sense and avoid" drones, augmented reality, autonomous vehicles, automated surveillance, and medical diagnostics. Differentiation is critical in these markets, where systems must have the highest response speed, the latest algorithms, and fast sensor deployment, and approximately two-thirds of vision-oriented semiconductor applications fall into this category.

     ReVISION supports building the most responsive vision system at the fastest speed. Compared with the most competitive computing embedded Gpus and typical SOCs, Revision improves the image capture speed per unit power inferred by machine learning by 6 times, improves the frame speed per unit power processed by computer vision by 42 times, and reduces the delay by 1/5. Even developers without hardware expertise can develop embedded vision applications with a single Zynq SoC or MPSoC chip using a combination of C/C++/OpenCL development processes, industry-standard frameworks, and libraries such as Caffe and OpenCV.

     With the unique reconfigurability and arbitrary connectivity benefits of the reVISION stack, developers can leverage the stack to quickly develop and deploy upgrades. As neural networks, algorithms, and sensor technologies and interface standards continue to accelerate, reconfigurability is critical to a "future-proof" intelligent vision system.

     The Xilinx reVISION stack includes rich development resources for platform, algorithm, and application development to support the most popular neural networks, including AlexNet, GoogLeNet, SqueezeNet, SSD, and FCN. In addition, the stack provides library elements, including a predefined optimized implementation of the CNN network layer, which is also required to build a custom neural network (DNN/CNN). Machine learning elements are combined with the rich OpenCV family of capabilities for computer vision processing to meet the requirements of acceleration. For application-level development, Xilinx supports industry-standard frameworks, including Caffe for machine learning and OpenVX for computer vision. The reVISION stack also includes development platforms provided by Xilinx and third parties, such as various types of sensors.

     Robert Chappell, CEO and founder of Eyetech Digital Systems, said: "Our eye tracking technology uses Zynq SoC to support high-definition visual analysis, which benefits patients with various types of paralysis, such as ALS. The new reVISION stack opens up a variety of new opportunities for algorithm development with the power of machine learning. This will enable us to further expand our human-computer interaction hardware offerings and enhance the value of our core eye tracking product."

     Lakshmi Mandyam, Senior Director of segment Marketing at ARM, said: "The embedded market is an evolving application area where changes in algorithms, neural networks and sensors require reconfiguration of target platforms. The ARM-based Zynq technology from Xilinx enables efficient deployment of these applications while accelerating widespread adoption of innovative machine learning applications from end to cloud."

     Steve Glaser, senior vice president of Strategy and Marketing at Xilinx, said: "We are seeing significant interest in machine learning applications from end to cloud, and we believe that Xilinx's continued investment in stack development will accelerate mainstream adoption. Today, hundreds of embedded vision customers have achieved more than 10x performance and latency improvements with Xilinx technology. With the new reVISION, thousands of customers will also benefit from these advantages."