Citation

Lim, Xin Roy and Wong, Lai Kuan and Loh, Yuen Peng and Gu, Ke and Lin, Weisi (2024) Mix-YOLONet: Deep Image Dehazing for Improving Object Detection. In: International Conference on Multimedia Modeling 2025, 9-11 January 2025, Nara, Japan.

Full text not available from this repository.

Abstract

Atmospheric haze significantly impairs the performance of computer vision tasks such as image dehazing and object detection. Existing methods often address these tasks independently, failing to provide an integrated solution that can effectively handle hazy conditions while maintaining accurate object detection. This paper presents Mix-YOLONet, a novel joint network architecture designed to tackle both image dehazing and object detection simultaneously. Mix-YOLONet leverages the powerful image restoration capabilities of U-Net-like architecture and integrates it with the detection precision of a YOLO head. Additionally, we integrated Mix Structure Blocks (MSB) to our joint network and experimented various configurations at strategic locations to enhance feature extraction and context aggregation. Through extensive experiments, we demonstrate that Mix-YOLONet achieves superior performance in both dehazing and object detection tasks under challenging visibility conditions, outperforming state-of-the-art methods on three benchmark datasets both quantitatively and qualitatively. The proposed joint network not only improves the clarity of hazy images but also ensures accurate object localization, paving the way for more robust object detection in adverse environmental conditions.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Object detection
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	18 Feb 2025 01:40
Last Modified:	18 Feb 2025 04:07
URII:	http://shdl.mmu.edu.my/id/eprint/13461

Downloads

Edit (login required)