Joint C2f and Joint Loss Object Detection Based on YOLOv5

Jinglin Cao

2024

Abstract

Seeking more accurate object detection in densely populated scenes, especially those involving small objects, is crucial for the development of computer vision applications. This study aims to significantly improve the detection capability of the YOLOv5 architecture. Specifically, this article proposes a new combination of C2f modules for enriching feature learning and distributed focus loss with a Complete Union Intersection (CIoU) loss function for improving object localization and class imbalance handling. Specifically, the C2f module helps to achieve better gradient propagation within the network, while the Distributed Focus Loss (DFL)+CIoU loss function improves detection accuracy through advanced boundary box calculations. This study was conducted on the COCO128 dataset. Its rigorous experimental framework confirms that these enhancements significantly improve average accuracy, recall, and bounding box accuracy. Experimental results indicate that the modified YOLOv5 model outperforms the baseline, offering significant improvements in detecting small-scale objects amidst complex backgrounds. The implications of this study are far-reaching, providing a foundation for developing real-time detection systems that are more reliable and effective across varied and challenging visual environments.

Paper Citation

in Harvard Style

Cao J. (2024). Joint C2f and Joint Loss Object Detection Based on YOLOv5. In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI; ISBN 978-989-758-713-9, SciTePress, pages 452-457. DOI: 10.5220/0012950800004508

in Bibtex Style

@conference{emiti24,
author={Jinglin Cao},
title={Joint C2f and Joint Loss Object Detection Based on YOLOv5},
booktitle={Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI},
year={2024},
pages={452-457},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012950800004508},
isbn={978-989-758-713-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI
TI - Joint C2f and Joint Loss Object Detection Based on YOLOv5
SN - 978-989-758-713-9
AU - Cao J.
PY - 2024
SP - 452
EP - 457
DO - 10.5220/0012950800004508
PB - SciTePress