Long-tailed instance segmentation using Gumbel optimized loss
File(s)
Accepted version
Author(s)
Alexandridis, Konstantinos Panagiotis
Deng, Jiankang
Nguyen, Anh
Luo, Shan
Type
Conference Paper
Abstract
Major advancements have been made in the field of object detection and segmentation recently. However, when it comes to rare categories, the state-of-the-art methods fail to detect them, resulting in a significant performance gap between rare and frequent categories. In this paper, we identify that Sigmoid or Softmax functions used in deep detectors are a major reason for low performance and are sub-optimal for long-tailed detection and segmentation. To address this, we develop a Gumbel Optimized Loss (GOL), for long-tailed detection and segmentation. It aligns with the Gumbel distribution of rare classes in imbalanced datasets, considering the fact that most classes in long-tailed detection have low expected probability. The proposed GOL significantly outperforms the best state-of-the-art method by 1.1% on AP, and boosts the overall segmentation by 9.0% and detection by 8.0%, particularly improving detection of rare classes by 20.3% , compared to Mask-RCNN, on LVIS dataset. Code available at: https://github.com/kostas1515/GOL.
Editor(s)
Avidan, S
Brostow, G
Cisse, M
Farinella, GM
Hassner, T
Date Issued
2022-11-03
Date Acceptance
2022-10-23
Citation
Computer Vision – ECCV 2022, 2022, 13670, pp.353-369
ISBN
978-3-031-20079-3
ISSN
0302-9743
Publisher
Springer International Publishing AG
Start Page
353
End Page
369
Journal / Book Title
Computer Vision – ECCV 2022
Volume
13670
Copyright Statement
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG. This version of the article has been accepted for publication, after peer review (when applicable) and is subject to Springer Nature’s AM terms of use, but is not the Version of Record and does not reflect post-acceptance improvements, or any corrections. The Version of Record is available online at: http://dx.doi.org/10.1007/978-3-031-20080-9_21
Source
17th European Conference on Computer Vision (ECCV)
Subjects
Computer Science
Computer Science, Artificial Intelligence
Gumbel activation
Imaging Science & Photographic Technology
Long-tailed distribution
Long-tailed instance segmentation
Science & Technology
Technology
Publication Status
Published
Start Date
2022-10-23
Finish Date
2022-10-27
Coverage Spatial
Tel Aviv, Israel
Date Publish Online
2023-11-03