This research investigates uncertainty, robustness, and data augmentation in deep learning, focusing on computer vision and generative AI. It examines the impact of augmentation on model calibration, evaluates automated techniques, and proposes a novel approach using pre trained diffusion models. The study aims to address limitations in current methods, improve model resilience, and enhance adaptability to real-world scenarios. By exploring the relationship between augmentation strategies and model performance, this research advances understanding of data augmentation’s influence on deep learning models in computer vision applications.