The new IP's capabilities include real-time 3D depth map generation and point cloud processing for 3D scanning. In addition, it can analyze scene information using the most processing-intensive object detection and recognition algorithms, ranging from ORB, Haar, and LBP, all the way to deep learning algorithms that use neural network technologies such as convolutional neural networks (CNN). The architecture also features a number of unique mechanisms, such as parallel random memory access and a patented two-dimension data processing scheme. These enable 4096-bit processing -- in a single cycle -- while keeping the memory bandwidth under 512bits for optimum energy efficiency.
In comparison to today's most advanced GPU cluster (points to NVIDIA, I guess), a single CEVA-XM4 core will complete a typical 'object detection and tracking' use-case scenario while consuming approximately 10% of the power and requiring approximately 5% of the die area.
Taking computer vision one step closer to human vision, the CEVA-XM4 also supports a wide range of computational photography algorithms that enhance the video or image, including refocus, background replacement, zoom, super-resolution, image stabilization, noise reduction and improved low-light capabilities.
