lens-ai/adversarial_pcam

对抗性PCAM数据集包含使用各种攻击技术生成的对抗性样本，这些样本旨在欺骗经过微调的模型。该数据集可用于评估模型对对抗攻击的鲁棒性，训练模型以改善其对对抗数据的弹性，以及基准新的对抗防御机制。数据集包含黑盒和非黑盒对抗攻击，并提供了每种攻击的成功率。

The Adversarial PCAM Dataset contains adversarial examples crafted using various attack techniques to fool a fine-tuned model. It is used for evaluating model robustness against adversarial attacks, training models with adversarial data for improved resilience, and benchmarking new adversarial defense mechanisms. The dataset includes both black-box and non-black-box adversarial attacks and provides success rates for each attack type.