Exploring Automated Data Augmentation Approaches for Deep Learning: A Case Study of Individual Feral Cat Classification
Zihan Yang, Richard Sinnott, James A Bailey, Krista A. Ehinger · Proceedings of the ... Annual Hawaii International Conference on System Sciences/Proceedings of the Annual Hawaii International Conference on System Sciences · 2024
This paper evaluates the performance of several automated data augmentation (AutoDA) methods for image classification problems suited for scenarios with limited and potentially imbalanced data sets.We compare one-stage, two-stage and search-free methods.These are explored in the context of a case study to identify/count feral cats in rural Victoria.Our results show that a trade-off exists between accuracy and efficiency, with one-stage methods being faster but less accurate than two-stage methods.Search-free methods are fastest, but have limited improvement in the resultant classification accuracy.