Computer Vision - ECCV 2022 - 17th European Conference, Tel Aviv, Israel, October 23-27, 2022, Proceedings, Part X
Computer Vision - ECCV 2022 - 17th European Conference, Tel Aviv, Israel, October 23-27, 2022, Proceedings, Part X
复制标题
计算机视觉 - ECCV 2022 - 第 17 届欧洲会议，以色列特拉维夫，2022 年 10 月 23-27 日，会议记录，第 X 部分
DOI：
10.1007/978-3-031-20080-9_21
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
Alexandridis K
中科院分区：
文献类型：
--
作者：
Alexandridis K
近年来，目标检测和分割领域取得了重大进展。然而，当涉及到稀有类别时，最先进的方法无法检测到它们，导致稀有类别和频繁类别之间存在显著的性能差距。在本文中，我们发现Sigmoid或Softmax函数用于深度检测器是导致性能低下的主要原因，并且对于长尾检测和分割是次优的。为了解决这一问题，我们提出了一种用于长尾检测和分割的Gumbel优化损失(GOL)。考虑到长尾检测中的大多数类的预期概率都很低，它与不平衡数据集中稀有类的Gumbel分布相一致。提出的GOL在AP上的性能明显优于目前最好的方法，并通过在LVIS数据集上通过与MASK-RCNN相比改进稀有类的检测来提高整体分割和检测的性能。代码请访问：https://github.com/kostas1515/GOL.
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 proposedGOLsignificantly outperforms the best state-of-the-art method byonAP, and boosts the overall segmentation byand detection by, particularly improving detection of rare classes by, compared to Mask-RCNN, on LVIS dataset. Code available at: https://github.com/kostas1515/GOL.