Revealing the Two Sides of Data Augmentation: An Asymmetric Distillation-based Win-Win Solution for Open-Set Recognition

TL;DR

This paper investigates the dual effects of data augmentation on open-set recognition, revealing challenges and proposing an asymmetric distillation approach with strategies to improve open-set performance, achieving state-of-the-art results.

Contribution

It introduces an asymmetric distillation framework with mutual information loss and relabeling strategies to balance closed-set and open-set recognition performance.

Findings

Reduces the decline in open-set recognition caused by data augmentation.

Achieves 2-3% higher AUROC on Tiny-ImageNet compared to SOTA.

Demonstrates generalization on large-scale ImageNet-21K.

Abstract

In this paper, we reveal the two sides of data augmentation: enhancements in closed-set recognition correlate with a significant decrease in open-set recognition. Through empirical investigation, we find that multi-sample-based augmentations would contribute to reducing feature discrimination, thereby diminishing the open-set criteria. Although knowledge distillation could impair the feature via imitation, the mixed feature with ambiguous semantics hinders the distillation. To this end, we propose an asymmetric distillation framework by feeding teacher model extra raw data to enlarge the benefit of teacher. Moreover, a joint mutual information loss and a selective relabel strategy are utilized to alleviate the influence of hard mixed samples. Our method successfully mitigates the decline in open-set and outperforms SOTAs by 2%~3% AUROC on the Tiny-ImageNet dataset and experiments on large-scale dataset ImageNet-21K demonstrate the generalization of our method.