Novel Class Discovery for Long-tailed Recognition
Abstract: While the novel class discovery has recently made great progress, existing methods typically focus on improving algorithms on class-balanced benchmarks. However, in real-world recognition tasks, the class distributions of their corresponding datasets are often imbalanced, which leads to serious performance degeneration of those methods. In this paper, we consider a more realistic setting for novel class discovery where the distributions of novel and known classes are long-tailed. One main challenge of this new problem is to discover imbalanced novel classes with the help of long-tailed known classes. To tackle this problem, we propose an adaptive self-labeling strategy based on an equiangular prototype representation of classes. Our method infers high-quality pseudo-labels for the novel classes by solving a relaxed optimal transport problem and effectively mitigates the class biases in learning the known and novel classes. We perform extensive experiments on CIFAR100, ImageNet100, Herbarium19 and large-scale iNaturalist18 datasets, and the results demonstrate the superiority of our method. Our code is available at \url{https://github.com/kleinzcy/NCDLR}.
Submission Length: Regular submission (no more than 12 pages of main content)
Previous TMLR Submission Url: https://openreview.net/forum?id=r8MzCT8zVD
Changes Since Last Submission: We have incorporated the feedback from the reviewers into the manuscript. This includes enhancing the clarity of our novelty in relation to existing work, conducting additional experiments on a larger dataset to demonstrate scalability, and providing a comprehensive analysis of our approach's performance in scenarios involving unknown novel classes.
Code: https://github.com/kleinzcy/NCDLR
Assigned Action Editor: ~Wei_Liu3
License: Creative Commons Attribution 4.0 International (CC BY 4.0)
Submission Number: 1088
Loading