Energy Drain of the Object Detection Processing Pipeline for Mobile Devices: Analysis and Implications

Applying deep learning to object detection provides the capability accurately detect and classify complex objects in real world. However, currently, few mobile applications use because such technology is computation-intensive energy-consuming. This article, best of our knowledge, presents first detailed experimental study a augmented reality (AR) client's energy consumption latency executing Convolutional Neural Networks (CNN) based detection, either locally on smartphone or remotely an edge server. In order measure obtain breakdown consumed by each phase processing pipeline, we propose new measurement strategy. Our measurements refine analysis AR clients reveal several interesting perspectives regarding CNN-based detection. Furthermore, insights research opportunities are proposed results. These findings from will guide design energy-efficient pipeline