Diverse dataset for image deterioration attacks

The dataset used in this paper is a diverse dataset created with real and synthetic images demonstrating that by exploiting spatio-temporal information combined with multi-component loss we significantly increase robustness against adverse image effects