Improving Calibration for Long-Tailed Recognition

TLDR

Deep neural networks often perform poorly on heavily class‑imbalanced datasets, and although two‑stage methods decouple representation and classifier learning, miscalibration remains a critical issue. The authors propose two methods to improve calibration and performance in long‑tailed recognition, specifically label‑aware smoothing to address over‑confidence varying with class frequency. They implement label‑aware smoothing and introduce shifted batch normalization within the decoupling framework to mitigate dataset bias between the two stages. The resulting approach sets new records on CIFAR‑10‑LT, CIFAR‑100‑LT, ImageNet‑LT, Places‑LT, and iNaturalist 2018, and the code will be released at https://github.com/Jia-Research-Lab/MiSLAS.

Abstract

Deep neural networks may perform poorly when training datasets are heavily class-imbalanced. Recently, two-stage methods decouple representation learning and classifier learning to improve performance. But there is still the vital issue of miscalibration. To address it, we design two methods to improve calibration and performance in such scenarios. Motivated by the fact that predicted probability distributions of classes are highly related to the numbers of class instances, we propose label-aware smoothing to deal with different degrees of over-confidence for classes and improve classifier learning. For dataset bias between these two stages due to different samplers, we further propose shifted batch normalization in the decoupling framework. Our proposed methods set new records on multiple popular long-tailed recognition benchmark datasets, including CIFAR-10-LT, CIFAR-100-LT, ImageNet-LT, Places-LT, and iNaturalist 2018. Code will be available at https://github.com/Jia-Research-Lab/MiSLAS.