Equalization Loss v2: A New Gradient Balance Approach for Long-tailed Object Detection

TLDR

Decoupled training methods dominate long‑tailed object detection, yet they require an extra fine‑tuning stage and can suffer from suboptimal representation–classifier optimization, while end‑to‑end approaches such as equalization loss still lag behind. This paper identifies that the main challenge in long‑tailed detection is the imbalance between positive and negative gradients, which existing EQL fails to address. To remedy this, the authors propose equalization loss v2, a gradient‑guided re‑weighting scheme that balances training for each category independently and equally. EQL v2 improves overall AP by about 4 points over the original EQL, gains 14–18 points on rare categories, surpasses decoupled methods, and boosts Open Images AP by 7.3 points without further tuning. Code is available at https://github.com/tztztztztz/eqlv2.

Abstract

Recently proposed decoupled training methods emerge as a dominant paradigm for long-tailed object detection. But they require an extra fine-tuning stage, and the dis-jointed optimization of representation and classifier might lead to suboptimal results. However, end-to-end training methods, like equalization loss (EQL), still perform worse than decoupled training methods. In this paper, we re-veal the main issue in long-tailed object detection is the imbalanced gradients between positives and negatives, and find that EQL does not solve it well. To address the problem of imbalanced gradients, we introduce a new version of equalization loss, called equalization loss v2 (EQL v2), a novel gradient guided reweighing mechanism that re-balances the training process for each category independently and equally. Extensive experiments are performed on the challenging LVIS benchmark. EQL v2 outperforms origin EQL by about 4 points overall AP with 14 ∼ 18 points improvements on the rare categories. More importantly, it also surpasses decoupled training methods. With-out further tuning for the Open Images dataset, EQL v2 improves EQL by 7.3 points AP, showing strong generalization ability. Codes have been released at https://github.com/tztztztztz/eqlv2