Dithering Defense: Adversarial Robustness of Vision Foundation Models via Multi-Level Floyd-Steinberg Dithering
The paper discusses the use of multi-level Floyd-Steinberg dithering to enhance the adversarial robustness of vision foundation models. This method serves as a lightweight, model-agnostic transformation that effectively disrupts adversarial attacks while maintaining semantic integrity. The authors demonstrate its effectiveness across various tasks and model families, outperforming existing techniques with minimal impact on clean inputs.
- ▪Vision foundation models are vulnerable to adversarial attacks, making them a critical point of failure.
- ▪The study evaluates multi-level Floyd-Steinberg dithering across six tasks and two model families.
- ▪Results indicate that this dithering method, especially when combined with post-processing blur, outperforms traditional baselines.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computer Vision and Pattern Recognition arXiv:2605.23065 (cs) [Submitted on 21 May 2026] Title:Dithering Defense: Adversarial Robustness of Vision Foundation Models via Multi-Level Floyd-Steinberg Dithering Authors:Yury Belousov, Brian Pulfer, Vitaliy Kinakh, Slava Voloshynovskiy View a PDF of the paper titled Dithering Defense: Adversarial Robustness of Vision Foundation Models via Multi-Level Floyd-Steinberg Dithering, by Yury Belousov and 2 other authors View PDF HTML (experimental) Abstract:Vision foundation models are widely used as frozen backbones across many downstream tasks, making them a single point of failure under adversarial attack.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
