Reproduce “ACTIVE: Towards Highly Transferable 3D Physical Camouflage for Universal and Robust Vehicle Evasion” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Naufal Suryanto, Yongsu Kim, Harashta Tatimma Larasati, Hyoeun Kang, Thi-Thu-Huong Le, Yoonyoung Hong, Hunmin Yang, Se-Yoon Oh and 1 more

The ACTIVE framework from Pusan National University develops highly transferable 3D physical adversarial camouflage that makes vehicles undetectable by state-of-the-art object detectors. This approach achieves significant performance drops (e.g., YOLOv3 [email protected] reduced to 1.28% in simulation and 8.75% in real-world tests) across diverse models, computer vision tasks, and physical environments.

Key insights

  • Object-independent texture mapping, such as Triplanar Mapping, is crucial for achieving broad universality of physical adversarial camouflage across different vehicle instances and classes.
  • Optimizing for complete undetectability through a novel Stealth Loss that drives both objectness and class confidence to zero is a more potent attack goal than merely misclassifying an object.
  • Combining physically realistic differentiable rendering with digital output augmentation significantly enhances the robustness of adversarial camouflage against diverse real-world conditions and state-of-the-art detection models.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.