Structural Knowledge Distillation for Object Detection

TL;DR

This paper introduces a novel structural similarity-based loss function for knowledge distillation in object detection, outperforming traditional lp-norm methods with minimal computational overhead.

Contribution

It replaces pixel-wise lp-norm with SSIM for better capturing structural cues, leading to significant performance improvements in object detection models.

Findings

Outperforms lp-norm based KD methods in object detection tasks.

Achieves +3.5 AP gain on Faster R-CNN R-50 over vanilla model.

Demonstrates effectiveness across various architectures and training schemes.

Abstract

Knowledge Distillation (KD) is a well-known training paradigm in deep neural networks where knowledge acquired by a large teacher model is transferred to a small student. KD has proven to be an effective technique to significantly improve the student's performance for various tasks including object detection. As such, KD techniques mostly rely on guidance at the intermediate feature level, which is typically implemented by minimizing an lp-norm distance between teacher and student activations during training. In this paper, we propose a replacement for the pixel-wise independent lp-norm based on the structural similarity (SSIM). By taking into account additional contrast and structural cues, feature importance, correlation and spatial dependence in the feature space are considered in the loss formulation. Extensive experiments on MSCOCO demonstrate the effectiveness of our method across different training schemes and architectures. Our method adds only little computational overhead, is straightforward to implement and at the same time it significantly outperforms the standard lp-norms. Moreover, more complex state-of-the-art KD methods using attention-based sampling mechanisms are outperformed, including a +3.5 AP gain using a Faster R-CNN R-50 compared to a vanilla model.