Investigating the effectiveness of data augmentation from similarity and diversity: An empirical study

Open Webpage

Suorong Yang, Suhan Guo, Jian Zhao, Furao Shen

Published: 01 Jan 2024, Last Modified: 15 May 2025Pattern Recognit. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•We propose quantitative measures to investigate the effectiveness of DA methods.•Our quantitative measures formulate the similarity and diversity metrics for DA.•The proposed measures are conducted in feature space, rather than raw pixel space.•Our measures reveals that most of best methods concentrate in a particular region.•Our study provides a comprehensive understanding of the mechanisms behind DA.