Summary of Dynamicpae: Generating Scene-aware Physical Adversarial Examples in Real-time, by Jin Hu et al.
DynamicPAE: Generating Scene-Aware Physical Adversarial Examples in Real-Time
by Jin Hu, Xianglong Liu, Jiakai Wang, Junkai Zhang, Xianqi Yang, Haotong Qin, Yuqing Ma, Ke Xu
First submitted to arxiv on: 11 Dec 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper presents DynamicPAE, a novel generative framework for physical adversarial examples (PAEs) that enables real-time attacks beyond static attacks. The key challenges in generating dynamic PAEs are exploring patterns under noisy gradient feedback and adapting to agnostic scenario natures. To address these issues, the authors introduce two techniques: residual-driven sample trajectory guidance and context-aligned scene expectation simulation process. The former helps break the limited feedback information restriction that leads to degeneracy problems, while the latter enhances robustness in incomplete observation contexts and facilitates consistent stealth control across different attack targets. Experimental results demonstrate DynamicPAE’s superior attack performance on object detectors like Yolo-v8, achieving a 1.95x boost (65.55% average AP drop under attack) over state-of-the-art static PAE generating methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research paper is about creating fake data that can trick artificial intelligence (AI) systems. The goal is to make the AI systems more robust and better at recognizing real objects. The authors developed a new method called DynamicPAE, which allows for real-time attacks on AI systems. They also introduced two new techniques to help their approach work better. The results show that their method is much better than other approaches in attacking AI systems. |
Keywords
» Artificial intelligence » Yolo
