Revisiting Knowledge Distillation for Object Detection

TL;DR

This paper introduces a flexible knowledge distillation framework for object detection that leverages pseudo labels, enabling use of unlabeled data, multiple teachers, and less labeled data, with demonstrated performance improvements.

Contribution

It proposes a novel distillation approach that relaxes the need for ground-truth labels, allowing for unlabeled data and multiple teachers, enhancing object detection performance.

Findings

Improves detection accuracy over existing methods.

Enables training with only 20% labeled data without performance loss.

Facilitates domain adaptation and multi-teacher integration.

Abstract

The existing solutions for object detection distillation rely on the availability of both a teacher model and ground-truth labels. We propose a new perspective to relax this constraint. In our framework, a student is first trained with pseudo labels generated by the teacher, and then fine-tuned using labeled data, if any available. Extensive experiments demonstrate improvements over existing object detection distillation algorithms. In addition, decoupling the teacher and ground-truth distillation in this framework provides interesting properties such: as 1) using unlabeled data to further improve the student's performance, 2) combining multiple teacher models of different architectures, even with different object categories, and 3) reducing the need for labeled data (with only 20% of COCO labels, this method achieves the same performance as the model trained on the entire set of labels). Furthermore, a by-product of this approach is the potential usage for domain adaptation. We verify these properties through extensive experiments.