A Unified Evaluation Framework of Physical Adversarial Robustness in the Edge VLM Era

Summary

This research benchmarks the physical adversarial robustness of compact grounding Vision-Language Models (VLMs) against object detectors, using a unified metric, conditional attack success rate (cASR), on images a model still localizes under a matched gray occluder. The authors split each failure into omission, substitution, and displacement, and find that VLMs can be as vulnerable as detectors. Failure mode tracks output format more than architecture, and under distribution shift VLMs that mainly omitted often shift toward displacement or substitution, which return confident, well-formed boxes.

Topics

• Adversarial machine learning

Related items

• InfoWhere AI Vulnerability Discovery Harnesses Go Quiet: Structural False Negatives and Hidden VulnerabilitiesSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoCorrelated Failures of AI Approval MonitorsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoRobust Fine‐Tuning Based on Fuzz Testing for Chinese BERT‐Based PLMsSimilar attack · OpenAlex (peer-reviewed AI security)
• InfoA Unified Attack Framework against Statistical Generative Watermarks for Large Language ModelsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoExposing Vulnerability Monoculture Through Systemic Adversarial RobustnessSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)