A Comprehensive Survey on Data Augmentation
Zaitian Wang, Pengfei Wang, Kunpeng Liu, Pengyang Wang, Yanjie Fu, Chang-Tien Lu, Charu C. Aggarwal, Jian Pei, Yuanchun Zhou
TL;DR
This work delivers a modality-independent, data-centric taxonomy for data augmentation, unifying methods across image, text, graph, tabular, and time-series data. It introduces a two-tier framework that distinguishes single-, multi-, and dataset-level augmentation and analyzes how value and structure information are leveraged within and across modalities. The paper catalogs a wide range of methods—from pixel-level perturbations to diffusion-based sample generation and exogenous data usage—while discussing evaluation metrics, target selection, and cross-modality transfer. By surveying up-to-date literature and offering guidance on method selection and future directions, it aims to standardize understanding of augmentation practices and promote cross-domain innovations.
Abstract
Data augmentation is a series of techniques that generate high-quality artificial data by manipulating existing data samples. By leveraging data augmentation techniques, AI models can achieve significantly improved applicability in tasks involving scarce or imbalanced datasets, thereby substantially enhancing AI models' generalization capabilities. Existing literature surveys only focus on a certain type of specific modality data and categorize these methods from modality-specific and operation-centric perspectives, which lacks a consistent summary of data augmentation methods across multiple modalities and limits the comprehension of how existing data samples serve the data augmentation process. To bridge this gap, this survey proposes a more enlightening taxonomy that encompasses data augmentation techniques for different common data modalities by investigating how to take advantage of the intrinsic relationship between and within instances. Additionally, it categorizes data augmentation methods across five data modalities through a unified inductive approach.