研究人员开发了TsallisPGD，一种新颖的对抗性攻击方法，旨在更有效地针对语义分割模型。这种新方法利用Tsallis交叉熵（标准交叉熵的广义形式）来适应性地调整跨像素的梯度加权。在Cityscapes和Pascal VOC等数据集上的实验表明，TsallisPGD在降低模型准确性和平均交并比（mIoU）方面优于现有方法。 AI
影响 引入了更强大的攻击向量来评估语义分割模型的鲁棒性。
排序理由 这是一篇研究论文，详细介绍了一种用于语义分割模型对抗性攻击的新方法。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2605.03405v1 Announce Type: new Abstract: Attacking semantic segmentation models is significantly harder than image classification models because an attacker must flip thousands of pixel predictions simultaneously. Standard pixel-wise cross-entropy (CE) is ill-suited to thi…
Attacking semantic segmentation models is significantly harder than image classification models because an attacker must flip thousands of pixel predictions simultaneously. Standard pixel-wise cross-entropy (CE) is ill-suited to this setting: it tends to overemphasize already-mis…