72e8efe6c1acc40f0b4a672a4330813e1a4066e5,art/attacks/evasion/auto_attack.py,AutoAttack,_run_attack,#AutoAttack#Any#Any#Any#Any#,189

Before Change

        // Check and update successful examples
        norm_is_smaller_eps = (
            np.linalg.norm((x_robust_adv - x_robust).reshape((x_robust_adv.shape[0], -1)), axis=1, ord=self.norm)
            <= self.eps
        )

After Change

        ) <= self.eps

        if attack.targeted:
            samples_misclassified = np.argmax(y_pred_robust_adv, axis=1) == np.argmax(y_robust, axis=1)
        elif not attack.targeted:
            samples_misclassified = np.argmax(y_pred_robust_adv, axis=1) != np.argmax(y_robust, axis=1)
        else:
            raise ValueError

        sample_is_not_robust = np.logical_and(samples_misclassified, norm_is_smaller_eps)

        x_robust[sample_is_not_robust] = x_robust_adv[sample_is_not_robust]
        x[sample_is_robust] = x_robust

Instances