Risk-Bounded Distribution Reconstruction: Stable Statistic Calibration for Long-Tailed Recognition

Abstract

Long-tailed recognition suffers from extreme class imbalance, where scarce tail data leads to biased and fragile feature distributions that exacerbate confusion with semantically or visually similar classes. Prior feature-space reconstruction methods transfer head-class structure or train conditional generators to synthesize tail features, yet the resulting statistical updates are often heuristic and can degrade multi-class separability when tail estimates are unreliable. Given this issue, we propose Risk-Bounded Distribution Reconstruction (RBDR), an offline statistic calibration framework for the two-stage long-tailed pipeline, grounded in an analysis of rival-induced discriminative directions. RBDR performs (i) risk-aware mean calibration by softly projecting any candidate update onto a supportive set such that a surrogate discriminative-risk upper bound does not increase, and (ii) covariance control by shrinking dispersion in a rival subspace while preserving diversity in orthogonal components. These plug-and-play modules transform heuristic reconstruction signals into controllable updates, improving performance and stability across multiple long-tailed benchmarks.