arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.10639cs.CVcs.CRcs.MM
收藏

VideoSTF: 视频大语言模型中输出重复性压力测试

VideoSTF: Stress-Testing Output Repetition in Video Large Language Models

National University of Singapore(新加坡国立大学);University of New South Wales(新南威尔士大学);CSIRO’s Data61(CSIRO数据61)

Yuxin Cao, Wei Song, Shangzhi Xu, Jingling Xue, Jin Song Dong

AI总结:

VideoSTF通过系统测量和压力测试揭示视频大语言模型中普遍存在的输出重复问题,发现其对时间扰动高度敏感,并暴露了其作为安全漏洞的风险。

摘要:

视频大语言模型(VideoLLMs)最近在视频理解任务中取得了强劲的性能。然而,我们发现了一种此前未被充分探索的生成失败模式:严重的输出重复,其中模型退化为自我强化的重复短语或句子的循环。这种失败模式未被现有VideoLLM基准所捕捉,这些基准主要关注任务准确性和事实正确性。我们引入VideoSTF,这是首个系统测量和压力测试VideoLLMs输出重复性的框架。VideoSTF使用三种互补的n-gram指标来形式化重复,并提供一个包含10000种多样视频的标准测试环境,以及一个受控的时间变换库。使用VideoSTF,我们在10种先进的VideoLLMs上进行了广泛测试、时间压力测试和对抗性利用。我们发现输出重复普遍存在,并且关键的是,对视频输入的时间扰动高度敏感。此外,我们展示了简单的时序变换可以高效地在黑盒环境下诱导重复退化,揭示输出重复作为可利用的安全漏洞。我们的结果揭示了输出重复作为现代VideoLLMs中的基本稳定性问题,并促使对视频-语言系统的稳定性意识评估。我们的评估代码和脚本可在:https://github.com/yuxincao22/VideoSTF_benchmark获取。

英文摘要:

Video Large Language Models (VideoLLMs) have achieved strong performance on video understanding tasks, yet existing benchmarks evaluate only what models predict, leaving the stability of how they generate largely unexamined. We surface a previously underexplored generation failure of VideoLLMs, defined as output repetition, in which the decoder collapses into self-reinforcing loops of repeated phrases or sentences, and present VideoSTF, a benchmarking framework for systematically measuring, stress-testing, and exploiting this failure mode. VideoSTF formalizes repetition with three complementary $n$-gram-based metrics, ships a standardized testbed of 10,000 diverse videos, and provides a library of controlled temporal stressors. Across 10 advanced VideoLLMs, VideoSTF reveals four key findings: (i) repetition is pervasive on unperturbed videos and stable across commonly used frame counts, with repetition rates up to 91%; (ii) it spans a severity spectrum from mild redundancy to token-cap loops, and is highly amplified by temporal perturbations; (iii) temporal stressors form a practical black-box attack surface, flipping benign videos into repetitive ones with tens of queries and high attack success rates (up to 98%), and (iv) repetition is not explained by visual redundancy, its amplification tracks local temporal disruption, and only repetition penalties reduce it among common mitigations such as top-$k$ sampling, input filtering, and prompt variation, but increasing the penalty weakens visual grounding. VideoSTF reframes generation stability as a useful and complementary evaluation axis for VideoLLMs and provides the tools to study it. The project page is available at https://videostf.github.io/.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑