Toward Highly Efficient Semantic-Guided Machine Vision for Low-Light Object Detection


Xin Feng (Chongqing University of Technology), Junxian Zeng (Chongqing University of Technology), Siping Wang (Chongqing University of Technology), Zhenwei He (Chongqing University of Technology)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)

Abstract

Detectors trained on well-lit data often experience significant performance degradation when applied to low-light conditions. To address this challenge, low-light enhancement methods are commonly employed to improve detection performance. However, existing human vision-oriented enhancement methods have shown limited effectiveness, which overlooks the semantic information for detection and achieves high computation costs. To overcome these limitations, we introduce a machine vision-oriented highly efficient low-light object detection method with the Efficient semantic-guided Machine Vision-oriented module (EMV). EMV can dynamically adapt to the object detection part based on end-to-end training and emphasize the semantic information for the detection. Besides, by lightening the network for feature decomposition and generating the enhanced image on latent space, EMV is a highly lightweight network for image enhancement, which contains only 27K parameters and achieves high inference speed. Extensive experiments conducted on ExDark and DarkFace datasets demonstrate that our method significantly improves detector performance in low-light environments. Our code is now available at https://github.com/Zeng555/EMV-YOLO.

Citation

@inproceedings{Feng_2024_BMVC,
author    = {Xin Feng and Junxian Zeng and Siping Wang and Zhenwei He},
title     = {Toward Highly Efficient Semantic-Guided Machine Vision for Low-Light Object Detection},
booktitle = {35th British Machine Vision Conference 2024, {BMVC} 2024, Glasgow, UK, November 25-28, 2024},
publisher = {BMVA},
year      = {2024},
url       = {https://papers.bmvc2024.org/0262.pdf}
}


Copyright © 2024 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection