Dual Attention Suppression Attack: Generate Adversarial Camouflage in Physical World

Abstract

Deep learning models are vulnerable to adversarial examples. As a more threatening type for practical deep learning systems, physical adversarial examples have received extensive research attention in recent years. However, without exploiting the intrinsic characteristics such as model-agnostic and human-specific patterns, existing works generate weak adversarial perturbations in the physical world, which fall short of attacking across different models and show visually suspicious appearance. Motivated by the viewpoint that attention reflects the intrinsic characteristics of the recognition process, this paper proposes the Dual Attention Suppression (DAS) attack to generate visuallynatural physical adversarial camouflages with strong transferability by suppressing both model and human attention. As for attacking, we generate transferable adversarial camouflages by distracting the model-shared similar attention patterns from the target to non-target regions. Meanwhile, based on the fact that human visual attention always focuses on salient items (e.g., suspicious distortions), we evade the human-specific bottom-up attention to generate visuallynatural camouflages which are correlated to the scenario context. We conduct extensive experiments in both the digital and physical world for classification and detection tasks on up to date models (e.g., Yolo-V5) and significantly demonstrate that our method outperforms state-of-the-art methods. 1 * Corresponding author 1 Our code can be found in https://github.com/nlsdesafety-team/DualAttentionAttack.

DPACFuse: Dual-Branch Progressive Learning for Infrared and Visible Image Fusion with Complementary Self-Attention and Convolution

“…We represent the feature extraction of the process in two stages. The intermediate features {F 2 ir , F 2 vi } obtained in the first stage can be expressed by the following formula:…”

“…Infrared and visible image fusion (IVF) has gained significant attention and is widely used in various applications [ 1 ]. Specifically, the effective fusion of shared and specific features from different modalities enables the generation of high-quality fused images, which, in turn, benefits downstream tasks, such as object detection [ 2 , 3 , 4 ], medical image processing [ 5 , 6 ], semantic segmentation [ 7 , 8 , 9 ], and pedestrian detection [ 10 , 11 ]. Although IVF has received much attention in various applications, IVF remains challenging due to the significant differences in appearance between these two image types.…”

Towards Accurate Post-Training Quantization for Vision Transformer

“…With the development of deep learning, the neural networks achieve great success in a various domains, such as image classification [24,37,39,[42][43][44], object detection [13,14,26,33,36,45], semantic segmentation [11,51], etc. Recently, the Vision Transformer (ViT) [9] emerges as a novel and effective architecture and shows great potential for various vision tasks.…”