Search papers, labs, and topics across Lattice.
This paper introduces adversarial Voronoi camouflage, which optimizes seed-point locations to create structured patterns that evade detection in security-critical systems. By employing a soft assignment method with fixed, printable palettes, the approach enhances visual plausibility while maintaining robustness against various detection models, as evidenced by significant performance drops in garment-level applications. The findings reveal that the camouflage can transfer effectively across different backgrounds and detector families, although repainting with different palettes can largely nullify its effectiveness.
Adversarial Voronoi camouflage achieves robust evasion against multiple detection models while maintaining visual plausibility, challenging traditional adversarial patch designs.
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.
