Technique Report of CVPR 2024 PBDL Challenges

Ying Fu,Yu Li,Shaodi You,Boxin Shi,Linwei Chen,Yunhao Zou,Zichun Wang,Yichen Li,Yuze Han,Yingkai Zhang,Jianan Wang,Qinglin Liu,Wei Yu,Xiaoqian Lv,Jianing Li,Shengping Zhang,Xiangyang Ji,Yuanpei Chen,Yuhan Zhang,Weihang Peng,Liwen Zhang,Zhe Xu,Dingyong Gou,Cong Li,Senyan Xu,Yunkang Zhang,Siyuan Jiang,Xiaoqiang Lu,Licheng Jiao,Fang Liu,Xu Liu,Lingling Li,Wenping Ma,Shuyuan Yang,Haiyang Xie,Jian Zhao,Shihua Huang,Peng Cheng,Xi Shen,Zheng Wang,Shuai An,Caizhi Zhu,Xuelong Li,Tao Zhang,Liang Li,Yu Liu,Chenggang Yan,Gengchen Zhang,Linyan Jiang,Bingyi Song,Zhuoyu An,Haibo Lei,Qing Luo,Jie Song,Yuan Liu,Qihang Li,Haoyuan Zhang,Lingfeng Wang,Wei Chen,Aling Luo,Cheng Li,Jun Cao,Shu Chen,Zifei Dou,Xinyu Liu,Jing Zhang,Kexin Zhang,Yuting Yang,Xuejian Gou,Qinliang Wang,Yang Liu,Shizhan Zhao,Yanzhao Zhang,Libo Yan,Yuwei Guo,Guoxin Li,Qiong Gao,Chenyue Che,Long Sun,Xiang Chen,Hao Li,Jinshan Pan,Chuanlong Xie,Hongming Chen,Mingrui Li,Tianchen Deng,Jingwei Huang,Yufeng Li,Fei Wan,Bingxin Xu,Jian Cheng,Hongzhe Liu,Cheng Xu,Yuxiang Zou,Weiguo Pan,Songyin Dai,Sen Jia,Junpei Zhang,Puhua Chen,Qihang Li

2024-07-12

Abstract:The intersection of physics-based vision and deep learning presents an exciting frontier for advancing computer vision technologies. By leveraging the principles of physics to inform and enhance deep learning models, we can develop more robust and accurate vision systems. Physics-based vision aims to invert the processes to recover scene properties such as shape, reflectance, light distribution, and medium properties from images. In recent years, deep learning has shown promising improvements for various vision tasks, and when combined with physics-based vision, these approaches can enhance the robustness and accuracy of vision systems. This technical report summarizes the outcomes of the Physics-Based Vision Meets Deep Learning (PBDL) 2024 challenge, held in CVPR 2024 workshop. The challenge consisted of eight tracks, focusing on Low-Light Enhancement and Detection as well as High Dynamic Range (HDR) Imaging. This report details the objectives, methodologies, and results of each track, highlighting the top-performing solutions and their innovative approaches.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper presents the results of the "Physics-based Vision and Deep Learning" (PBDL) challenge held at the CVPR 2024 conference. The challenge consists of 8 tracks that focus on low-light enhancement and detection as well as high-dynamic-range (HDR) imaging. These tasks aim to improve the robustness and accuracy of computer vision techniques in complex environments by utilizing the combination of physical principles and deep learning methods.

The low-light enhancement and detection track includes tasks such as low-light object detection and instance segmentation, low-light original video denoising, low-light SRGB image enhancement, extreme low-light image denoising, and low-light original image enhancement. The objective is to improve the image quality and visual effects under low-light conditions, especially for areas such as autonomous driving and surveillance.

The high-dynamic-range imaging track involves HDR reconstruction of single original images, event-based high-speed HDR video reconstruction, and overexposure correction of original images. The goal is to capture a wider range of brightness and enhance the realism and details of the images.

The challenge attracted multiple teams from around the world, each proposing innovative methods to address these problems. The report provides detailed descriptions of the objectives, methods, and results of each track, highlighting outstanding solutions. Through this competition, researchers demonstrated the potential of combining physics-based vision and deep learning in low-light processing and HDR imaging, promoting advancements in related fields.