Novel Class Discovery for Long-tailed Recognition
This addresses a realistic challenge in real-world recognition tasks where class imbalance is common, though it is incremental as it adapts existing methods to a more specific setting.
The paper tackles the problem of novel class discovery under long-tailed class distributions, where both known and novel classes are imbalanced, and proposes an adaptive self-labeling strategy with equiangular prototypes to mitigate class biases, achieving superior results on datasets like CIFAR100 and iNaturalist18.
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 https://github.com/kleinzcy/NCDLR.