Improving Transferability of Adversarial Examples With Input Diversity
Though CNNs have achieved the state-of-the-art performance on various vision tasks, they are vulnerable to adversarial examples --- crafted by adding human-imperceptible perturb...