Existing attacks almost perform on low-level instances, such as pixels and super-pixels, and rarely exploit semantic clues.

Our RADIAL-RL agents consistently outperform prior methods when tested against attacks of varying strength and are more computationally efficient to train. In addition, we propose a new evaluation method calledGreedyWorst-Case Reward(GWC) tomeasure attack agnostic robustness of deep RL agents. We show that GWC can be evaluated efficiently and is a good estimate of the reward under the worst possible sequence of adversarial attacks.

Internal learning for single-image generation is a framework where a generatoristrained toproduce novelimages based on asingle image. Since these modelsare trained on asingle image, theyare limited in their scale and application.