Activate and Reject: Towards Safe Domain Generalization under Category Shift

被引:2
|
作者
Chen, Chaoqi [1 ]
Tang, Luyao [2 ]
Tao, Leitian [3 ]
Zhou, Hong-Yu [1 ]
Huang, Yue [2 ]
Han, Xiaoguang [4 ]
Yu, Yizhou [1 ]
机构
[1] Univ Hong Kong, Hong Kong, Peoples R China
[2] Xiamen Univ, Xiamen, Peoples R China
[3] Univ Wisconsin, Madison, WI USA
[4] Chinese Univ Hong Kong, Shenzhen, Peoples R China
来源
2023 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2023) | 2023年
关键词
D O I
10.1109/ICCV51070.2023.01061
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Albeit the notable performance on in-domain test points, it is non-trivial for deep neural networks to attain satisfactory accuracy when deploying in the open world, where novel domains and object classes often occur. In this paper, we study a practical problem of Domain Generalization under Category Shift (DGCS), which aims to simultaneously detect unknown-class samples and classify known-class samples in the target domains. Compared to prior DG works, we face two new challenges: 1) how to learn the concept of "unknown" during training with only source known-class samples, and 2) how to adapt the source-trained model to unseen environments for safe model deployment. To this end, we propose a novel Activate and Reject (ART) framework to reshape the model's decision boundary to accommodate unknown classes and conduct post hoc modification to further discriminate known and unknown classes using unlabeled test data. Specifically, during training, we promote the response to the unknown by optimizing the unknown probability and then smoothing the overall output to mitigate the overconfidence issue. At test time, we introduce a step-wise online adaptation method that predicts the label by virtue of the cross-domain nearest neighbor and class prototype information without updating the network's parameters or using threshold-based mechanisms. Experiments reveal that ART consistently improves the generalization capability of deep networks on different vision tasks. For image classification, ART improves the H-score by 6.1% on average compared to the previous best method. For object detection and semantic segmentation, we establish new benchmarks and achieve competitive performance.
引用
收藏
页码:11518 / 11529
页数:12
相关论文
共 50 条
  • [1] Upcycling Models under Domain and Category Shift
    Qu, Sanqing
    Zou, Tianpei
    Roehrbein, Florian
    Lu, Cewu
    Chen, Guang
    Tao, Dacheng
    Jiang, Changjun
    2023 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2023, : 20019 - 20028
  • [2] Forecasting battery degradation trajectory under domain shift with domain generalization
    Tan, Ruifeng
    Lu, Xibin
    Cheng, Minhao
    Li, Jia
    Huang, Jiaqiang
    Zhang, Tong-Yi
    ENERGY STORAGE MATERIALS, 2024, 72
  • [3] Curriculum learning-based domain generalization for cross-domain fault diagnosis with category shift
    Wang, Yu
    Gao, Jie
    Wang, Wei
    Yang, Xu
    Du, Jinsong
    MECHANICAL SYSTEMS AND SIGNAL PROCESSING, 2024, 212
  • [4] CodeS: Towards Code Model Generalization Under Distribution Shift
    Hu, Qiang
    Guo, Yuejun
    Xie, Xiaofei
    Cordy, Maxime
    Papadakis, Mike
    Ma, Lei
    Le Traon, Yves
    2023 IEEE/ACM 45TH INTERNATIONAL CONFERENCE ON SOFTWARE ENGINEERING-NEW IDEAS AND EMERGING RESULTS, ICSE-NIER, 2023, : 1 - 6
  • [6] Towards Unsupervised Domain Generalization
    Zhang, Xingxuan
    Zhou, Linjun
    Xu, Renzhe
    Cui, Peng
    Shen, Zheyan
    Liu, Haoxin
    2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022), 2022, : 4900 - 4910
  • [7] Diversity-Based Generalization for Unsupervised Text Classification Under Domain Shift
    Krishnan, Jitin
    Purohit, Hemant
    Rangwala, Huzefa
    MACHINE LEARNING AND KNOWLEDGE DISCOVERY IN DATABASES, ECML PKDD 2020, PT II, 2021, 12458 : 657 - 672
  • [8] Towards Improving Calibration in Object Detection Under Domain Shift
    Munir, Muhammad Akhtar
    Khan, Muhammad Haris
    Sarfraz, M. Saquib
    Ali, Mohsen
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 35 (NEURIPS 2022), 2022,
  • [9] Category-Stitch Learning for Union Domain Generalization
    Liu, Yajing
    Xiong, Zhiwei
    Li, Ya
    Lu, Yuning
    Tian, Xinmei
    Zha, Zheng-Jun
    ACM TRANSACTIONS ON MULTIMEDIA COMPUTING COMMUNICATIONS AND APPLICATIONS, 2023, 19 (01)
  • [10] Towards Principled Disentanglement for Domain Generalization
    Zhang, Hanlin
    Zhang, Yi-Fan
    Liu, Weiyang
    Weller, Adrian
    Schoelkopf, Bernhard
    Xing, Eric P.
    2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2022, : 8014 - 8024