论文标题
动态分裂和诱使对抗性训练可用于稳健语义分段
Dynamic Divide-and-Conquer Adversarial Training for Robust Semantic Segmentation
论文作者
论文摘要
对抗性训练有望提高深层神经网络对对抗性扰动的鲁棒性，尤其是在分类任务上。相反，这种类型的培训对语义分割的影响只是开始的。我们通过制定一般的对抗性训练程序，可以在对抗性和干净样本上表现出色，从而初步尝试探索语义细分的防御策略。我们提出了一种动态的分裂和诱使对抗训练（DDC-AT）策略，以增强防御效应，通过在训练过程中设置目标模型中的其他分支，并处理具有不同特性的像素对对抗性扰动的处理。我们的动态分裂机制将像素自动分为多个分支。请注意，所有这些其他分支可以在推理过程中放弃，因此没有额外的参数和计算成本。在Pascal VOC 2012和CityScapes数据集上进行了各种细分模型的广泛实验，其中DDC-AT在白色和黑色盒子攻击下都能产生令人满意的性能。
Adversarial training is promising for improving robustness of deep neural networks towards adversarial perturbations, especially on the classification task. The effect of this type of training on semantic segmentation, contrarily, just commences. We make the initial attempt to explore the defense strategy on semantic segmentation by formulating a general adversarial training procedure that can perform decently on both adversarial and clean samples. We propose a dynamic divide-and-conquer adversarial training (DDC-AT) strategy to enhance the defense effect, by setting additional branches in the target model during training, and dealing with pixels with diverse properties towards adversarial perturbation. Our dynamical division mechanism divides pixels into multiple branches automatically. Note all these additional branches can be abandoned during inference and thus leave no extra parameter and computation cost. Extensive experiments with various segmentation models are conducted on PASCAL VOC 2012 and Cityscapes datasets, in which DDC-AT yields satisfying performance under both white- and black-box attack.
