CVPR 2024 PBDL 挑战赛技术报告
计算机视觉与模式识别
2024-07-15 v3
摘要
基于物理的视觉与深度学习的交叉为推进计算机视觉技术提供了一个令人兴奋的前沿领域。通过利用物理原理来指导并增强深度学习模型，我们可以开发出更鲁棒、更准确的视觉系统。基于物理的视觉旨在逆转过程，从图像中恢复场景属性，如形状、反射率、光照分布和介质属性。近年来，深度学习在各种视觉任务中展现出有希望的改进，当与基于物理的视觉相结合时，这些方法可以增强视觉系统的鲁棒性和准确性。本技术报告总结了在 CVPR 2024 研讨会上举办的“基于物理的视觉遇见深度学习”(PBDL) 2024 挑战赛的成果。该挑战赛包含八个赛道，聚焦于低光照增强与检测以及高动态范围(HDR)成像。本报告详细介绍了每个赛道的目标、方法和结果，重点介绍了表现最佳的解决方案及其创新方法。
引用
@article{arxiv.2406.10744,
title = {Technique Report of CVPR 2024 PBDL Challenges},
author = {Ying Fu and Yu Li and Shaodi You and Boxin Shi and Linwei Chen and Yunhao Zou and Zichun Wang and Yichen Li and Yuze Han and Yingkai Zhang and Jianan Wang and Qinglin Liu and Wei Yu and Xiaoqian Lv and Jianing Li and Shengping Zhang and Xiangyang Ji and Yuanpei Chen and Yuhan Zhang and Weihang Peng and Liwen Zhang and Zhe Xu and Dingyong Gou and Cong Li and Senyan Xu and Yunkang Zhang and Siyuan Jiang and Xiaoqiang Lu and Licheng Jiao and Fang Liu and Xu Liu and Lingling Li and Wenping Ma and Shuyuan Yang and Haiyang Xie and Jian Zhao and Shihua Huang and Peng Cheng and Xi Shen and Zheng Wang and Shuai An and Caizhi Zhu and Xuelong Li and Tao Zhang and Liang Li and Yu Liu and Chenggang Yan and Gengchen Zhang and Linyan Jiang and Bingyi Song and Zhuoyu An and Haibo Lei and Qing Luo and Jie Song and Yuan Liu and Qihang Li and Haoyuan Zhang and Lingfeng Wang and Wei Chen and Aling Luo and Cheng Li and Jun Cao and Shu Chen and Zifei Dou and Xinyu Liu and Jing Zhang and Kexin Zhang and Yuting Yang and Xuejian Gou and Qinliang Wang and Yang Liu and Shizhan Zhao and Yanzhao Zhang and Libo Yan and Yuwei Guo and Guoxin Li and Qiong Gao and Chenyue Che and Long Sun and Xiang Chen and Hao Li and Jinshan Pan and Chuanlong Xie and Hongming Chen and Mingrui Li and Tianchen Deng and Jingwei Huang and Yufeng Li and Fei Wan and Bingxin Xu and Jian Cheng and Hongzhe Liu and Cheng Xu and Yuxiang Zou and Weiguo Pan and Songyin Dai and Sen Jia and Junpei Zhang and Puhua Chen and Qihang Li},
journal= {arXiv preprint arXiv:2406.10744},
year = {2024}
}
备注
CVPR 2024 PBDL Challenges: https://pbdl-ws.github.io/pbdl2024/challenge/index.html