What is it about?

Real-time object detection powered by deep learning is critical for Smart Transport and Smart City. Edge computing combined with the cloud can minimize latency, but efficient machine resource management is necessary. While deep learning-based object detection provides accurate and reliable results, it demands high computational power. This reveals the need to implement models with less complex architectures for edge deployment. This study evaluates the performance of evolving deep learning models and their lightweight versions, including YOLOv5-Nano, YOLOX-Nano, YOLOX-Tiny, YOLOv6-Nano, YOLOv6-Tiny, and YOLOv7-Tiny, on a commercially available edge device. The results show that YOLOv5-Nano and YOLOv6-Nano with their TensorRT versions can provide real-time applicability in approximately 35 milliseconds of inference time. Additionally, YOLOv6-Tiny achieves the highest average precision, while YOLOv5-Nano exhibits the lowest energy consumption when compared to other models.

Featured Image

Why is it important?

This paper presents a performance analysis of the latest YOLO series models on the Jetson Nano edge device. Additionally, we compare the lightweight nano and tiny versions of various YOLO series object detectors to determine their suitability for different requirements.

Perspectives

This paper will be useful for researchers working on computer vision combined with edge computing area. It provides performance evaluation of state-of-the-art object detectors on a well-known edge device. It can also ease the determination of which model to use if for example, energy consumption is the main concern or the model accuracy, etc.

Anilcan Bulut
Marmara Universitesi

Read the Original

This page is a summary of: Performance Evaluation of Recent Object Detection Models for Traffic Safety Applications on Edge, January 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3582177.3582178.
You can read the full text:

Read

Contributors

The following have contributed to this page