AI and deep learning showcased at Vision 2018

Building upon Xilinx’s state-of-the-art DeePhi solutions for optimised and accelerated deep-learning, this demo showcases a system that offers breakthrough performance with energy efficiency combined with low latency and re-configurability which is required for edge inference solutions such as object detection, face recognition, segmentation and video analytics.

Xilinx Machine Learning Acceleration on Alveo enables developers to optimise and deploy accelerated machine-learning inference utilising Xilinx Alveo Data Center accelerator cards or cloud-based environments. The solution is based on the Xilinx xDNN (Deep Neural Network) processing engine – a high-performance energy-efficient DNN accelerator that outperforms many common CPU and GPU platforms.

The highly flexible Denali-MC HDR is a video/still image-processing platform from Pinnacle Imaging Systems. This system demo runs on a Xilinx Zynq 7000 FPGA with an ON Semiconductor AR0239 sensor and achieves a combination of highest possible dynamic range, while compensating for HDR motion artefacts.

Avnet Silica’s AI-accelerated multi-camera development platform speeds development of computer-vision, AI and machine-learning designs. The system offers a complete imaging solution connecting multiple cameras to a Xilinx Zynq UltraScale+ MPSoC development platform.

Targeting cloud-based surveillance applications this demo showcases Skreen’s Personalised Streaming Video Engine running on Xilinx Alveo accelerator cards, as well as on AWS EC2 and Nimbix NX5 cloud services.

Using a modular architecture, the Skreens Interactive Experience Engine uses patented Interactive Multi-Layering technology to deliver real-time, encoded video and interactive content to any output incorporating machine-learning overlays into video-intensive products and services.

The final demo is based around the XGS12000 image sensor from ON Semiconductor for industrial imaging applications.

Simplifying and accelerating time-to-market for new camera designs, the sensor can enable a single camera design that supports multiple resolutions as well as different pixel designs with only minor modifications being required.

“We are using Vision 2018 to showcase the very latest in next-generation AI and machine learning technology,” said Alexander Friebe, Supplier Line Manager at Avnet Silica. “With the unique end-to-end ecosystem, and working very closely with leading partners like Xilinx and ON Semiconductor, Avnet Silica is continually looking to bring the latest imaging technologies to customers building advanced solutions for industrial markets.”