Defense Against Adversarial Attacks on No-Reference Image Quality Models with Gradient Norm Regularization

Improving the adversarial robustness of No-Reference Image Quality Assessment models through gradient norm regularization.