D2R: dual regularization loss with collaborative adversarial generation for model robustness

Re-assign community