Journal of Applied Optics, Volume. 44, Issue 5, 1037(2023)
Nighttime low-light image enhancement and object detection based on knowledge distillation
Figures & Tables(7)
Fig. 1. Structure diagram of dual task model of image enhancement and object detection and schematic of knowledge distillation
Fig. 3. Visual comparison on low-light image enhancement effect of different methods
Fig. 4. Test results of mAP indicators on real low-light images dataset
Fig. 5. Test results of low-light image enhancement and object detection
Table 1. Parameters of image quality assessment of different methods
View tableView in ArticleTable 1. Parameters of image quality assessment of different methods
Method/Index NIQE PSNR SSIM EnlightenGAN[ 18 ]18.82 28.23 0.73 Zero-DCE[ 21 ]23.33 28.54 0.73 CycleGAN[41] 27.58 28.61 0.71 Algorithm proposed 19.86 28.57 0.74
Table 2. mAP of different methods
View tableView in ArticleTable 2. mAP of different methods
Method mAP50 mAP50@ExDark YOLOXtiny(Normal illumination) 80.47% None YOLOXtiny(LLL+Knowledge distillation) 54.40% 41.28% YOLOXtiny(LLL) 37.82% 23.13%
Tools
Get Citation
Copy Citation Text
Delin MIAO, Lei LIU, Yongchao MO, Chaolong HU, Yijun ZHANG, Yunsheng QIAN. Nighttime low-light image enhancement and object detection based on knowledge distillation[J]. Journal of Applied Optics, 2023, 44(5): 1037
Paper Information
Category: Research Articles
Received: Nov. 24, 2022
Accepted: --
Published Online: Mar. 12, 2024
The Author Email: LIU Lei (刘磊)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20