Domain Similarity-Perceived Label Assignment for Domain Generalized Underwater Object Detection

Authors: Xisheng Li, Wei Li∗, Pinhao Song, Mingjun Zhang, and Guifang Sun Xisheng Li. School of Artificial Intelligence and Computer Science, Jiangnan University, Jiangsu, P. R. China Email: 6213113079@stu.jiangnan.edu.cn Wei Li*. Corresponding author. School of Artificial Intelligence and Computer Science, Jiangnan University, Jiangsu, P. R. China. Email: cs weili@jiangnan.edu.cn Pinhao Song are with Robotics Research Group, the Department of Mechanical Engineering, KU Leuven, Belgium. Email: pinhao.song@kuleuven.be Mingjun Zhang. School of Artificial Intelligence and Computer Science Jiangnan University, Jiangsu, P. R. China. Email: mingjunzhang@stu.jiangnan.edu.cn Guifang Sun. School of Mechanical Engineering Southeast University, Nanjing, P. R. China. Email: gfsun@seu.edu.cn

Conference: ICIC 2024 Posters, Tianjin, China, August 5-8, 2024

Pages: 607-621

Keywords: Index Terms—Domain adversarial learning, underwater object detection, pseudo domain label.

DOI: 10.65286/icic.v20i1.14897

Abstract

Abstract—The inherent characteristics and light fluctuations of water bodies give rise to the huge difference between different layers and regions in underwater environments. When
the test set is collected in a different marine area from the training set, the issue of domain shift emerges, significantly compromising the model's ability to generalize. The Domain
Adversarial Learning DAL training strategy has been previously utilized to tackle such challenges. However, DAL heavily depends on manually one-hot domain labels, which implies
no difference among the samples in the same domain. Such an assumption results in the instability of DAL. This paper introduces the concept of Domain Similarity-Perceived Label
Assignment DSP . The domain label for each image is regarded as its similarity to the specified domains. Through domainspecific data augmentation techniques,
we achieved state-of-the-art results on the underwater cross-domain object detection benchmark S-UODAC2020. Furthermore, we validated the effectiveness of our method in the Cityscapes dataset.

BibTeX Citation:

@inproceedings{ICIC2024,
    author = {Xisheng Li, Wei Li∗, Pinhao Song, Mingjun Zhang, and Guifang Sun Xisheng Li. School of Artificial Intelligence and Computer Science, Jiangnan University, Jiangsu, P. R. China Email: 6213113079@stu.jiangnan.edu.cn Wei Li*. Corresponding author. School of Artificial Intelligence and Computer Science, Jiangnan University, Jiangsu, P. R. China. Email: cs weili@jiangnan.edu.cn Pinhao Song are with Robotics Research Group, the Department of Mechanical Engineering, KU Leuven, Belgium. Email: pinhao.song@kuleuven.be Mingjun Zhang. School of Artificial Intelligence and Computer Science Jiangnan University, Jiangsu, P. R. China. Email: mingjunzhang@stu.jiangnan.edu.cn Guifang Sun. School of Mechanical Engineering Southeast University, Nanjing, P. R. China. Email: gfsun@seu.edu.cn},
    title = {Domain Similarity-Perceived Label Assignment for Domain Generalized Underwater Object Detection},
    booktitle = {Proceedings of the 20th International Conference on Intelligent Computing (ICIC 2024)},
    month = {August},
    date = {5-8},
    year = {2024},
    address = {Tianjin, China},
    pages = {607-621},
    note = {Poster Volume Ⅰ}
    doi = {10.65286/icic.v20i1.14897}
}