Adaptive Resolution-Based Tradeoffs for Energy-Efficient Visual Computing Systems

The real world presents interpretable visual detail at different scales in different situations. While empowering face recognition, augmented reality, and other computer vision tasks, mobile systems should be able to dynamically adapt the spatiotemporal resolution of the visual sensing pipeline to capture image frames at high resolutions for task precision and low resolutions for energy savings. Facilitating real-time decisions to reconfigure resolutions will let systems dynamically adapt to the needs of the vision algorithms, as well as the environmental situation of the visual scene. This article will review system challenges and opportunities of image-resolution-based tradeoffs toward energy-efficient visual computing through device driver and media framework optimization.