aiqbench analysis
This research introduces a novel adversarial camouflage technique using Voronoi diagrams to degrade person detection models across various YOLO architectures without requiring pixel-wise optimization. By optimizing only seed-point locations under fixed palettes, the method achieves robust evasion in black-box settings while maintaining visual plausibility comparable to traditional splinter patterns. The study highlights significant vulnerabilities in current computer vision systems when facing structured geometric attacks that transfer effectively across different backgrounds and detector families.
Structured Adversarial Camouflage via Voronoi Diagrams
arXiv:2606.17711v1 Announce Type: cross Abstract: Pixel-wise adversarial patches are computationally heavy and often visually detectable, limiting utility in security-critical systems. We present adversarial Voronoi camouflage that optimizes only seed-point locations under fixed, printable palettes using a soft assignment, producing structured, splinter camouflage-like patterns without additional regularization. Evaluated on person detection with COCO-style AP@[.5:.95], naive placement (Inria -> COCO) performs comparably bad, while garment-level application via segmentation mask (3DPeople) results in a significant AP drop. The attack transfers to out-of-domain backgrounds and across detector families (YOLOv9/10/11/12), indicating robustness in black-box settings. Repainting with different palettes largely nullifies the effect, and single-color tweaks show limited tolerance (<=0.17), highlighting a structure-palette coupling. The parameter-efficient, palette-constrained design improves visual plausibility while degrading real-time detector performance. Physical validation and color calibration are left for future work. Code: https://github.com/JensBayer/Voronoi This paper was originally presented at the International Conference on Military Communication and Information Systems (ICMCIS), organized by the Information Systems Technology (IST) Scientific and Technical Committee, IST-224-RSY - the ICMCIS, held in Bath, United Kingdom, 12-13 May 2026.
Abstract:Pixel-wise adversarial patches are computationally heavy and often visually detectable, limiting utility in security-critical systems. We present adversarial Voronoi camouflage that optimizes only seed-point locations under fixed, printable palettes using a soft assignment, producing structured, splinter camouflage-like patterns without additional regularization. Evaluated on person detection with COCO-style AP@[.5:.95], naive placement (Inria -> COCO) performs comparably bad, while garment-level application via segmentation mask (3DPeople) results in a significant AP drop. The attack transfers to out-of-domain backgrounds and across detector families (YOLOv9/10/11/12), indicating robustness in black-box settings. Repainting with different palettes largely nullifies the effect, and single-color tweaks show limited tolerance (<=0.17), highlighting a structure-palette coupling. The parameter-efficient, palette-constrained design improves visual plausibility while degrading real-time detector performance. Physical validation and color calibration are left for future work. Code: this https URL This paper was originally presented at the International Conference on Military Communication and Information Systems (ICMCIS), organized by the Information Systems Technology (IST) Scientific and Technical Committee, IST-224-RSY - the ICMCIS, held in Bath, United Kingdom, 12-13 May 2026.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2606.17711 [cs.CV]
(or arXiv:2606.17711v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2606.17711 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 269 words
Related coverage
RL-ADA: A World-Feedback Framework for Adversarially Robust Enterprise Dialogue Agents
arXiv · Sep 4, 2026
MorphStrata: Layer-Specific Perturbations for Generating Morphence Students in Time-Series Moving Target Defense
arXiv · Jun 17, 2026
Auditor-in-a-Box: Tools for Third-Party Auditing
LessWrong · Jul 28, 2026
Abduction Without a Body? Representational Grounding and the Abduction Loop for Scientific Hypothesis Generation
arXiv · Aug 5, 2026