Concurrent Programmable Data Co-processor

Tagging images and videos on mobile devices is an open market. Many software companies recently began to work on deep learning techniques; however, such techniques are limited by the low-performance hardware of mobile phones. In addition, some of the existing products are task-specific and cannot be reprogramed to perform new tasks.

Researchers at Purdue University have developed a new system of hardware and software that can solve state-of-the-art, complex, vision tasks. This system includes a novel hardware architecture that provides low-power operation and simultaneously a high number of operations for a whole class of vision algorithms while maintaining complete programmability. This system adapts new tasks with minor reprogramming and scale more efficiently to new vision models.

Advantages:

-Can solve state-of-the-art, complex, vision tasks

-Provides low-power operation and simultaneously a high number of operations

-Adapt new tasks with minor reprogramming

-Scale more efficiently to new vision models

TRL: 6

Intellectual Property:

Keywords: mobile vision tasks, deep learning limitations, low-power hardware, high-number operations, novel hardware architecture, programmable vision system, efficient vision models, vision algorithms, mobile image tagging, reprogrammable tasks