Benchmark Analysis of YOLO Performance on Edge Intelligence Devices

Feng, Haogang; Mu, Gaoze; Zhong, Shida; Zhang, Peichang; Yuan, Tao

doi:10.1109/csrswtc52801.2021.9631594

Cited by 7 publications

(5 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, this study [35] aimed to learn about the inference workflow and performance of the YOLO network by using the edge devices of Jetson Nano, Jetson Xavier NX, and Raspberry Pi 4B with NCS. In this benchmark, two versions of the YOLO network were used to detect different video contents across the three edge devices mentioned above.…”

Section: Discussionmentioning

confidence: 99%

Implementation of a Cluster-Based Heterogeneous Edge Computing System for Resource Monitoring and Performance Evaluation

et al. 2022

View full text Add to dashboard Cite

In the past decade, Internet of Things (IoT) technology has been widely used in various applications in people's daily life. Currently, IoT applications mainly depend on the powerful cloud datacenters as the computing and storage centers. However, with such cloud-centric frameworks, a large amount of data will be transferred between the end devices and the remote cloud datacenters via a long wide-area network, which may potentially result in intolerable latency and a lot of energy consumption. To alleviate this problem, the edge computing (EC) paradigm is exploited to sink the cloud computing capability from the network core to network edges in proximity to end devices, so as to enable computationintensive and latency-critical edge intelligence applications to be executed in a real-time manner. With the increasing amount of edge devices, it is essential to obtain the status of devices in real-time for realizing the overall resources of heterogeneous edge devices. Thus, it is important to construct a system which can be used to monitor each device's status and obtain the performance of each device. In this study, a cluster-based heterogeneous edge computing environment is implemented for resource monitoring and performance evaluation. In the experiment, three deep learning models of object detection are used to evaluate the performance of the implemented system. Through the experimental results, we can easily realize the resource usage, including the high range dynamics of electricity and power consumption of heterogeneous edge devices via the visualization results.INDEX TERMS Edge computing, resource monitoring, Kubernetes, Prometheus, Grafana.

show abstract

Section: Discussionmentioning

confidence: 99%

Implementation of a Cluster-Based Heterogeneous Edge Computing System for Resource Monitoring and Performance Evaluation

et al. 2022

View full text Add to dashboard Cite

show abstract

“…We evaluated different embedded boards that can run the application in our use case and can be installed in a drone (given their power consumption, weight, size, and price). Nowadays two solutions stand out: 41 A low-power computing board, such as a Raspberry Pi, with a neural computing peripheral such as the Intel NCS 2; § § and specialised low-consumption boards integrating a GPU for machine learning acceleration, such as the Nvidia Jetson boards. ¶ ¶ In this regard, the Jetson Nano and Jetson Xavier NX boards are suitable, as they have been used in similar experiments.…”

Section: Onboard Versus Offload Computingmentioning

confidence: 99%

“…The study in Reference 41 compared the performances of an Intel NCS 2, a Jetson Nano and a Jetson Xavier NX for common AI applications. The Jetson boards can execute most general AI models.…”

Section: Experimental Evaluationmentioning

confidence: 99%

“…We verified that this was not due to high CPU usage nor CPU throttling, and memory usage was also well under maximum capacity. According to Reference 41, for the YOLOv3 model, which is equivalent in complexity to our application, the Jetson Nano has 0.64–0.68 times the performance of a Raspberry Pi 4 with an Intel NCS 2, whereas the Jetson Xavier NX outperforms the Intel NCS 2 by a 2.36–2.44 factor. We therefore estimate that the Jetson Nano and Jetson Xavier NX could respectively process 3 and 11 frames per second on average with AI models like ours in similar conditions.…”

Section: Experimental Evaluationmentioning

confidence: 99%

“…During offloading evaluations, the power consumption of the payload was only around 4.05 W. Of these, 0.97 W corresponded to the Logitech Streamcam web camera, which was also used for onboard computing. In Reference 41 the authors reported respective average power consumptions of 7.95 and 15.2 W for the Nvidia Jetson Nano and Xavier NX boards running a YOLOv3 object detection model. On the other hand, the DJI Matrice M210 drone in our tests has a reported flight time of 23 minutes without any payload using two regular DJI TB50 batteries.…”

Section: Experimental Evaluationmentioning

confidence: 99%

See 2 more Smart Citations

Is the edge really necessary for drone computing offloading? An experimental assessment in carrier‐grade 5G operator networks

et al. 2022

View full text Add to dashboard Cite

In this article, we evaluate the first experience of computation offloading from drones to real fifth‐generation (5G) operator systems, including commercial and private carrier‐grade 5G networks. A follow‐me drone service was implemented as a representative testbed of remote video analytics. In this application, an image of a person from a drone camera is processed at the edge, and image tracking displacements are translated into positioning commands that are sent back to the drone, so that the drone keeps the camera focused on the person at all times. The application is characterised to identify the processing and communication contributions to service delay. Then, we evaluate the latency of the application in a real non standalone 5G operator network, a standalone carrier‐grade 5G private network, and, to compare these results with previous research, a Wi‐Fi wireless local area network. We considered both multi‐access edge computing (MEC) and cloud offloading scenarios. Onboard computing was also evaluated to assess the trade‐offs with task offloading. The results determine the network configurations that are feasible for the follow‐me application use case depending on the mobility of the end user, and to what extent MEC is advantageous over a state‐of‐the‐art cloud service.

show abstract