Distilling Image Classifiers in Object Detectors

TL;DR

This paper introduces a novel classifier-to-detector knowledge distillation method that transfers knowledge across tasks, improving object detector performance by leveraging classification teachers, and outperforms existing detector-to-detector distillation techniques.

Contribution

The paper proposes a new framework for transferring knowledge from classifiers to detectors, enabling cross-task distillation and enhancing detection accuracy and localization.

Findings

Outperforms state-of-the-art detector-to-detector distillation methods

Effective across various detectors and backbones

Improves recognition accuracy and localization performance

Abstract

Knowledge distillation constitutes a simple yet effective way to improve the performance of a compact student network by exploiting the knowledge of a more powerful teacher. Nevertheless, the knowledge distillation literature remains limited to the scenario where the student and the teacher tackle the same task. Here, we investigate the problem of transferring knowledge not only across architectures but also across tasks. To this end, we study the case of object detection and, instead of following the standard detector-to-detector distillation approach, introduce a classifier-to-detector knowledge transfer framework. In particular, we propose strategies to exploit the classification teacher to improve both the detector's recognition accuracy and localization performance. Our experiments on several detectors with different backbones demonstrate the effectiveness of our approach, allowing us to outperform the state-of-the-art detector-to-detector distillation methods.