Task-Balanced Distillation for Object Detection

TL;DR

This paper introduces Task-Balanced Distillation, a method that improves object detection by aligning classification and regression tasks through harmony scoring and feature distillation, leading to better student model performance.

Contribution

The paper proposes a novel harmony score and task-decoupled feature distillation to address spatial misalignment in object detection knowledge distillation.

Findings

RetinaNet with ResNet-50 achieves 41.0 mAP on COCO with TBD.

TBD outperforms recent distillation methods like FGD and FRS.

The method demonstrates strong generalization across models and datasets.

Abstract

Mainstream object detectors are commonly constituted of two sub-tasks, including classification and regression tasks, implemented by two parallel heads. This classic design paradigm inevitably leads to inconsistent spatial distributions between classification score and localization quality (IOU). Therefore, this paper alleviates this misalignment in the view of knowledge distillation. First, we observe that the massive teacher achieves a higher proportion of harmonious predictions than the lightweight student. Based on this intriguing observation, a novel Harmony Score (HS) is devised to estimate the alignment of classification and regression qualities. HS models the relationship between two sub-tasks and is seen as prior knowledge to promote harmonious predictions for the student. Second, this spatial misalignment will result in inharmonious region selection when distilling features. To alleviate this problem, a novel Task-decoupled Feature Distillation (TFD) is proposed by flexibly balancing the contributions of classification and regression tasks. Eventually, HD and TFD constitute the proposed method, named Task-Balanced Distillation (TBD). Extensive experiments demonstrate the considerable potential and generalization of the proposed method. Specifically, when equipped with TBD, RetinaNet with ResNet-50 achieves 41.0 mAP under the COCO benchmark, outperforming the recent FGD and FRS.