🤖 AI Summary
This paper identifies a previously overlooked issue in long-tailed (LT) classification: attribute-level imbalance—where intra-class samples exhibit long-tailed distributions due to latent attributes (e.g., pose, illumination, occlusion), even when class labels are balanced; such attributes are combinatorially complex and hard to explicitly rebalance. To address this, we formulate *Generalized Long-Tailed Classification* (GLT), unifying inter-class and intra-class (attribute-level) imbalance modeling. Our contributions are threefold: (1) the first formal definition of GLT and two benchmarks—ImageNet-GLT and MSCOCO-GLT; (2) an environment-discovery-driven Invariant Feature Learning (IFL) paradigm that automatically identifies heterogeneous attribute environments via model prediction inconsistency, then jointly optimizes class-balancedness and attribute robustness through environment-adversarial learning and feature disentanglement; (3) IFL serves as a plug-and-play module, consistently boosting performance of diverse re-sampling, augmentation, and ensemble methods on both benchmarks.
📝 Abstract
Existing long-tailed classification (LT) methods only focus on tackling the class-wise imbalance that head classes have more samples than tail classes, but overlook the attribute-wise imbalance. In fact, even if the class is balanced, samples within each class may still be long-tailed due to the varying attributes. Note that the latter is fundamentally more ubiquitous and challenging than the former because attributes are not just implicit for most datasets, but also combinatorially complex, thus prohibitively expensive to be balanced. Therefore, we introduce a novel research problem: Generalized Long-Tailed classification (GLT), to jointly consider both kinds of imbalances. By"generalized", we mean that a GLT method should naturally solve the traditional LT, but not vice versa. Not surprisingly, we find that most class-wise LT methods degenerate in our proposed two benchmarks: ImageNet-GLT and MSCOCO-GLT. We argue that it is because they over-emphasize the adjustment of class distribution while neglecting to learn attribute-invariant features. To this end, we propose an Invariant Feature Learning (IFL) method as the first strong baseline for GLT. IFL first discovers environments with divergent intra-class distributions from the imperfect predictions and then learns invariant features across them. Promisingly, as an improved feature backbone, IFL boosts all the LT line-up: one/two-stage re-balance, augmentation, and ensemble. Codes and benchmarks are available on Github: https://github.com/KaihuaTang/Generalized-Long-Tailed-Benchmarks.pytorch
