Sunnyvale, CA (PRWEB) May 28, 2015
CNN-based object detection solutions have been shown to offer higher accuracy than conventional computer vision algorithms. However, Neural Networks require significantly more compute power to operate in real-time. Most current CNN-based object detection solutions are designed for powerful platforms with high-end CPUs and graphics processors. With their naturally parallel computing models, Neural Networks are well suited to implementation on FPGAs. Now with the SDAccel development environment, FPGAs can offer the advantages of high performance, re-configurability and fast development.
MulticoreWare develops and licenses a wide range of computer vision and video processing libraries, while also providing design services to Xilinx customers. High performance applications for FPGAs can be developed quickly and efficiently using the SDAccel development environment. Leveraging MulticoreWare’s libraries and services, Xilinx customers will now be able to deploy a variety of CNN-based object detection solutions for traffic monitoring, home security and situational awareness. In the broadcast industry, CNN-based solutions include video analysis and quality assurance applications including the detection of audio-video synchronization errors, video banding and video macroblocking.
“Our CNN-based face detection tool for the Alpha Data ADM-PCIE-7V3 board consists of 8 layers and is based on the Alexnet model. The face classifier was trained using the Torch7 framework. Training data consisted of 160,317 snapshots of both face and non-face images re-sized to 32X32 pixels, resulting in a training accuracy of around 95%. With the powerful capabilities of the SDAccel environment and Xilinx FPGAs, MulticoreWare is well positioned to develop machine learning related solutions for Xilinx customers,” said Curtis Davis, VP of Engineering at MulticoreWare Inc.
“MulticoreWare’s unique combination of capabilities in High Performance Computing, OpenCL programming, and deep expertise in the key application area of Neural Networks makes them a valued member of the SDAccel development environment-certified ecosystem,” said Tom Feist, senior director, Design Methodology Marketing, Xilinx. “This combination of expertise coupled with Xilinx’s tools and platforms provides an ideal solution to accelerate deployment of FPGA-based accelerators for video & image processing, neural networks and computer applications.
MulticoreWare, Inc. is a leading provider of high performance video processing software libraries including the x265 HEVC encoder, UHDcode HEVC decoder, GPU accelerated Video Processing Library and GPU accelerated VP9 decoders. MulticoreWare's parallel computing tools, libraries and expertise in heterogeneous computing forms the foundation of a worldwide professional services business, with more than 200 engineers in 6 locations. For more information on x265 please visit multicorewareinc.com, or contact sales @ multicorewareinc.com