Teacher Network Calibration Improves Cross-Quality Knowledge Distillation

TL;DR

This paper introduces cross-quality knowledge distillation (CQKD), where a high-resolution trained teacher guides a low-resolution student, improving classification accuracy and calibration, especially with temperature smoothing.

Contribution

It proposes CQKD as a method to transfer knowledge from high- to low-resolution networks, enhancing performance and calibration in image classification.

Findings

CQKD outperforms supervised learning in large-scale image classification.

Higher temperature smoothing improves student calibration and accuracy.

Calibrating the teacher's output distribution benefits student network performance.

Abstract

We investigate cross-quality knowledge distillation (CQKD), a knowledge distillation method where knowledge from a teacher network trained with full-resolution images is transferred to a student network that takes as input low-resolution images. As image size is a deciding factor for the computational load of computer vision applications, CQKD notably reduces the requirements by only using the student network at inference time. Our experimental results show that CQKD outperforms supervised learning in large-scale image classification problems. We also highlight the importance of calibrating neural networks: we show that with higher temperature smoothing of the teacher's output distribution, the student distribution exhibits a higher entropy, which leads to both, a lower calibration error and a higher network accuracy.