Decoupling Representation and Classifier for Long-Tailed Recognition

TL;DR

This paper proposes decoupling representation learning from classifier training to improve long-tailed recognition, demonstrating that simple, balanced representations combined with classifier adjustment outperform complex existing methods.

Contribution

It introduces a decoupled approach to long-tailed recognition, showing that high-quality representations can be learned independently and classifiers can be adjusted to handle class imbalance effectively.

Findings

Representation learning may not be hindered by class imbalance.

Adjusting classifiers alone can achieve strong long-tailed recognition.

State-of-the-art results on multiple benchmarks with a simple decoupled method.

Abstract

The long-tail distribution of the visual world poses great challenges for deep learning based classification models on how to handle the class imbalance problem. Existing solutions usually involve class-balancing strategies, e.g., by loss re-weighting, data re-sampling, or transfer learning from head- to tail-classes, but most of them adhere to the scheme of jointly learning representations and classifiers. In this work, we decouple the learning procedure into representation learning and classification, and systematically explore how different balancing strategies affect them for long-tailed recognition. The findings are surprising: (1) data imbalance might not be an issue in learning high-quality representations; (2) with representations learned with the simplest instance-balanced (natural) sampling, it is also possible to achieve strong long-tailed recognition ability by adjusting only the classifier. We conduct extensive experiments and set new state-of-the-art performance on common long-tailed benchmarks like ImageNet-LT, Places-LT and iNaturalist, showing that it is possible to outperform carefully designed losses, sampling strategies, even complex modules with memory, by using a straightforward approach that decouples representation and classification. Our code is available at https://github.com/facebookresearch/classifier-balancing.