Hardware Performance Evaluation of Different Computing Devices on YOLOv5 Ship Detection Model
Clint Aldrin A. Valencia, Rendell Sheen S. Suliva, Jocelyn F. Villaverde · 2022
Over the years, edge devices have a large impact in terms of processes for artificial intelligence. The advancement quickly created a more accessible technology for the development of AI. Different studies have applied their own ship detection model using a single device. This prevents the maximization of the performance of the model since the best hardware option is not determined. Therefore, in developing a custom detection model, it is necessary to determine the best hardware to be used in the deployment of the model. This study aims to determine the best hardware for deployment. The average pre-processing time is 33.86s, 50.33s, and 8.47s with an average FPS of 4.86, 1.25, and 27.2 for Jetson Nano, Raspberry Pi 4, and desktop computer, respectively. It is observed that the desktop computer performed significantly better, followed by Jetson Nano, and lastly, Raspberry Pi under the same input, configuration, and model. The deciding factor in these results are dependent on the component of the device, which are the OS, CPU, GPU, and memory.