CEVA, a maker of digital signal processors (DSPs) has launched its own real-time neural network software framework, called the CEVA Deep Neural Network (CDNN). Using its CEVA-XM4 DSP, CEVA claims that its CDNN enables embedded systems to carry out the necessary image interpretation three times faster than GPU-based alternatives. The CDNN claims that it can perform that 3x improvement while using 30x less power and needing 15x less memory bandwidth, with CEVA saying that running a CDNN-based pedestrian detection system in a car requires only 30mW for a 1080p video stream at 30fps – using a 28nm silicon architecture. CEVA says the system is designed for image and object recognition, advanced driver assistance systems (ADAS), AI, video analytics, augmented and…