Jeff Bier, President of Berkeley Design Technology, Inc. (BDTI) and Founder of the Embedded Vision Alliance, presents the "Choosing a Processor for Embedded Vision: Options and Trends" tutorial at the May 2015 Embedded Vision Summit.
Computer vision applications typically demand lots of processor performance. These applications also tend to be complex and fast-changing, so developers need processors that enable quick development of efficient code. And, as vision proliferates into new markets, more and more applications also require low cost and low power consumption. Taken together, this is a tough set of requirements for a processor. Processor suppliers have responded with a diverse range of architectures, combining CPUs with GPUs, DSPs, FPGAs, vision-specific processors and neural network engines.
In this presentation, Jeff provides a map to the landscape of processor options for vision applications, highlighting strengths and weaknesses of different processor types. He also illuminates important trends in processors for vision processors (such as specialized neural network engines) and associated development tools (such as support for standard APIs).