Introduction

• Python code for our paper (Highly differentiated target detection under extremely low-light conditions based on improved YOLOx model)
• Note: The original implementation of PyTorch version YOLOX, please vist: MegEngine implementation.

Updates

• 【2022/01/20】 The initial version is based on YOLOv3 for object detection under low-light conditions; more details are available at SSRN
• 【2023/10/21】 We optimized the training process using YOLOx with a larger dataset under low-light conditions, resulting in higher performance. See [notes](Coming soon) for more details.

TODO: Update with exact module version

numpy,torch>=1.7,opencv_python,loguru,tqdm,torchvision,thop,ninja,tabulate

Dataset

• Pascal VOC datasets from the VOC challenges are available through the challenge links: Link
• MS COCO 2017 datasets can be found at: Link

Benchmark

Table 1: Performance Comparison of Low-Light Image Detection

Methods	Proposed	YOLOx	YOLOv4	RFBnet	Mobilenet-SSD	Faster-RCNN	M2det
Dark image	70.96	67.41	54.96	64.06	53.91	65.19	65.09
Dong et al.[1]	72.15	68.47	47.75	64.09	36.52	62.71	62.20
This paper	76.86	73.31	69.83	73.79	63.26	62.75	72.32

Table 2: Analysis of Average Precision (AP) of YOLOx in Various Lighting Conditions

Image	AP@0.50:0.95	AP@0.50	AP@0.75	AP@S	AP@M	AP@L
Original image	0.504	0.690	0.547	0.325	0.561	0.669
Dark image	0.404	0.592	0.426	0.196	0.448	0.597
This paper	0.456	0.643	0.489	0.239	0.508	0.650

Table 3: Augmented Reality (AR) of YOLOx in Various Illumination Conditions

Image	AR@0.50:0.95	AR@0.50	AR@0.75	AR@S	AR@M	AR@L
Original image	0.379	0.614	0.653	0.468	0.712	0.825
Dark image	0.327	0.514	0.549	0.334	0.489	0.687
This paper	0.354	0.566	0.603	0.371	0.505	0.669

Reference

• [1] Xuan Dong et al., "Fast efficient algorithm for enhancement of low lighting video," 2011 IEEE International Conference on Multimedia and Expo, Barcelona, 2011, pp. 1-6, doi: 10.1109/ICME.2011.6012107.
• [2] Ge, Z., Liu, S., Wang, F., Li, Z., & Sun, J. (2021). Yolox: Exceeding yolo series in 2021. arXiv preprint arXiv:2107.08430.

Cite YOLOX

If you use YOLOX in your research, please cite our work by using the following BibTeX entry:

@article{shao2024highly,
  title={Highly differentiated target detection under extremely low-light conditions based on improved YOLOx model},
  author={Haijian Shao, Suqin Lei, Chenxu Yan, Xing Deng, Yunsong Qi},
  vol={140},
  no={2},
  pages={1507-1537},
  journal={CMES-Computer Modeling in Engineering & Sciences},
  doi={10.32604/cmes.2024.050140},
  year={2024}
}

Free Download: Link Feel free to quote articles from my Google Scholar profile: [Google Scholar]: Link