Increasing FPS for single board computers and embedded computers in 2021 (Jetson nano and YOVOv4-tiny). Practice and review

TL;DR

This paper reviews various methods to increase FPS on single-board computers, focusing on Nvidia Jetson and low-cost alternatives, and evaluates these methods using YOLOv4-tiny on Jetson Nano.

Contribution

It provides a comprehensive overview of techniques for boosting FPS on SBCs, including both GPU and CPU-based approaches, with experimental results on Jetson Nano.

Findings

Methods can significantly improve FPS on Jetson Nano.

GPU-based methods outperform CPU-only approaches.

Experimental results demonstrate practical improvements for YOLOv4-tiny.

Abstract

This manuscript provides a review of methods for increasing the frame per second of single-board computers. The main emphasis is on the Jetson family of single-board computers from Nvidia Company, due to the possibility of using a graphical interface for calculations. But taking into account the popular low-cost segment of single-board computers as RaspberryPI family, BananaPI, OrangePI, etc., we also provided an overview of methods for increasing the frame per second without using a Graphics Processing Unit. We considered frameworks, software development kit, and various libraries that can be used in the process of increasing the frame per second in single-board computers. Finally, we tested the presented methods for the YOLOv4-tiny model with a custom dataset on the Jetson nano and presented the results in the table.