Generalizing state-of-the-art object detectors for autonomous vehicles in unseen environments
作者：
Highlights：
• Different scenarios are used to generalize object detectors to new test domains.
• Adverse data scarcity for object detection in autonomous vehicle is addressed.
• The fragility of object detectors against different distortions is alleviated.
• Domain shift problem is mitigated through synthetic data generation.
• The impact of different strategies on cross-domain robustness is investigated.
摘要
•Different scenarios are used to generalize object detectors to new test domains.•Adverse data scarcity for object detection in autonomous vehicle is addressed.•The fragility of object detectors against different distortions is alleviated.•Domain shift problem is mitigated through synthetic data generation.•The impact of different strategies on cross-domain robustness is investigated.
论文关键词：Autonomous vehicles,Object detection,Distribution mismatch,Natural distortions,Cross-domain robustness
论文评审过程：Received 5 July 2020, Revised 15 May 2021, Accepted 9 June 2021, Available online 18 June 2021, Version of Record 22 June 2021.
论文官网地址：https://doi.org/10.1016/j.eswa.2021.115417
