DNN inference offloading for object detection in 5G multi-access edge computing
Geun-Yong Kim, Ryangsoo Kim, Sungchang Kim, Ki-Dong Nam, Sung-Uk Rha, Junghyun Yoon · 2021 International Conference on Information and Communication Technology Convergence (ICTC) · 2021
In this paper, we present experimental results of a Deep Neural Network (DNN) inference offloading in a realworld 5G MEC testbed to meet requirements for high detection accuracy and low end-to-end latency for object detection. Central Unit (CU), Distributed Unit (DU), Radio Unit (RU), and MEC servers are located in the 5G MEC testbed such that the performance of latency sensitive MEC applications could be tested. We implemented the DNN offloading by applying task pipeline parallelism and DNN task decoupling scheme from edge to MEC server. We verified the effectiveness of the DNN offloading using YOLOv3 in terms of Frames Per Second (FPS) and power consumption in the 5G MEC testbed.