You're running an image recognition project and realize that you do not have enough data of a certain type of vehicle. What is the best course of action to get the additional labeled data you need?

When class instances are scarce, image augmentation—geometric or photometric transformation that multiplies available samples—creates new labeled examples without additional manual effort, improving model generalization in vision tasks.

A. Buying data adds cost and may not match required format or license.
C. Sampling reshuffles existing data; it does not create new minority-class images.
D. Anonymization only removes identifying information; it produces no extra training samples.

1. Krizhevsky, A. et al., “ImageNet Classification with Deep CNNs”, NIPS 2012, §3 Data Augmentation.

2. Shorten, C. & Khoshgoftaar, T. M., “A Survey on Image Data Augmentation”, J. Big Data 6:60, 2019, §2.1. https://doi.org/10.1186/s40537-019-0197-0

3. Stanford CS231n Lecture 7, “Data Augmentation”, Spring 2022, slides 3-4.