City
Berlin
Country
Germany
Start Date
End Date
Abstract

<p>Join Cadence at the Concar Expo 2018 in booth 60, where we are presenting an ADAS reference platform featuring an SoC with four Cadence<sup>®</sup> Tensilica<sup>®</sup> Vision P6 DSPs&#046; The SoC achieves high&#8211;performance processing and high data throughput while keeping power consumption to a minimum&#046; This is especially important for image&#8211;processing applications in automotive, computer vision, and neural networks applications&#046;</p> <p>See our demos on:</p> <p><b>AI for pedestrian recognition</b><br>The On&#8211;Device Artificial Intelligence (AI) demonstration of pedestrian recognition uses the Tensilica Vision P6 DSP and TinyYOLO (You Only Look Once) algorithm&#046; The algorithm is based on a neural network and is trained in the recognition of persons&#046; TinyYOLO is ideal for applications that require fast, energy&#8211;efficient object detection (including localization) for multiple object categories&#046;</p> <p><b>AI for image classification and object recognition</b><br>Software developers use environments such as Caffe or TensorFlow in the design of neural networks for image classification&#046; The sometimes very complex networks then have to be manually ported to a hardware platform and optimized—a challenging and time&#8211;consuming task&#046; For this, Cadence has developed the Xtensa<sup>®</sup> Neural Network Compiler (XNNC), which shortens the time required to convert the neural network into code for a Tensilica AI DSP, as an embedded (target) processor, from months to just days&#046; An XNNC demonstration will show the implementation of a Caffe or TensorFlow&#8211;trained Google Inception V3 and MobileNet&#046; The XNNC converts them into a highly optimized fixed&#8211;point neural network code for the Tensilica Vision DSPs&#046;</p><p><br></p>