Dithering Defense: Adversarial Robustness of Vision Foundation Models via Multi-Level Floyd-Steinberg Dithering
arXiv:2605.23065v1 Announce Type: cross Abstract: Vision foundation models are widely used as frozen backbones across many downstream tasks, making them a single point of failure under adversarial attack. We study multi-level Floyd-Steinberg error-diffusion dithering as a lightweight, model-agnostic input transformation that disrupts adversarial perturbations while preserving semantic content. Unlike prior work, which was limited to binary dithering, grayscale CIFAR-10, and a single small model trained from scratch, we evaluate across six tasks (classification, segmentation, depth estimation,
The proliferation of vision foundation models across critical applications necessitates robust defenses against adversarial attacks, making this research timely.
Adversarial robustness is crucial for the reliability and trustworthiness of AI systems, especially as they integrate into high-stakes environments.
This research introduces a new, lightweight, and model-agnostic defense method that could significantly improve the security posture of vision foundation models across diverse tasks.
- · AI developers
- · Organizations deploying vision AI
- · Cybersecurity researchers
- · Adversarial attackers
- · Outdated adversarial defense methods
Vision foundation models become more resilient to adversarial perturbations, reducing immediate attack vectors.
Increased trust in AI deployments across sensitive applications like autonomous systems and critical infrastructure.
The arms race between AI security and adversarial attacks intensifies, driving further innovation in both areas.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG