Robust Adversarial Attack for Deep Neural Target Detection Network
Explored the adaptability of some classical methods including DPatch, DPAttack, and RPAttack on the DOTA dataset.
Introduced Class Patch Attack (CPAttack), a novel class-based geometric transformation robust adversarial method.
By targeting detection boxes for improved results, and directly applying perturbations based on box categories, CPAttack exhibits real-time applicability and practical significance.
Design and Implementation of a Quadruped Robot for AI Education