Mind the Temporal Gap: Evaluating Video Models for Temporal Understanding in Driving

Computer Science > Computer Vision and Pattern Recognition arXiv:2610.12575 (cs) [Submitted on 8 Oct 2026] Title: Mind the Temporal Gap: Evaluating Video Models for Temporal Understanding in Driving Authors: Shiva Sreeram, Erfan Aasi, Tsun-Hsuan Wang, Alaa Maalouf, Daniela Rus.

计算机科学 > 计算机视觉与模式识别 arXiv:2610.12575 (cs) [提交于 2026 年 10 月 8 日] 标题：关注时间鸿沟：评估用于驾驶时间理解的视频模型 作者：Shiva Sreeram, Erfan Aasi, Tsun-Hsuan Wang, Alaa Maalouf, Daniela Rus。

Abstract: Modern video-language models are increasingly treated as general-purpose video reasoners, yet it remains unclear whether they robustly capture temporal dynamics in safety-critical settings such as driving. We study this question through controlled ego-motion tasks that require identifying actions from driving videos.

摘要：现代视频语言模型正日益被视为通用视频推理工具，但尚不清楚它们是否能在驾驶等安全关键场景中稳健地捕捉时间动态。我们通过受控的自我运动任务研究了这一问题，这些任务要求从驾驶视频中识别动作。

Across video-language models, we find that standard prompting, visual prompting, few-shot prompting, and chain-of-thought prompting remain unreliable and highly task- and model-dependent, with persistent class biases, although newer models show substantial task-specific few-shot gains.

在各类视频语言模型中，我们发现标准提示、视觉提示、少样本提示和思维链提示仍然不可靠，且高度依赖于特定任务和模型，并存在持续的类别偏差，尽管较新的模型在特定任务的少样本学习中表现出了显著提升。

In contrast, lightweight LoRA finetuning and simple probes trained on intermediate visual features recover strong motion-discriminative performance, indicating that motion-relevant cues are often present inside the models but are not reliably used by the final video-to-language pathway.

相比之下，轻量级的 LoRA 微调和基于中间视觉特征训练的简单探测器恢复了强大的运动判别性能，这表明模型内部通常存在与运动相关的线索，但最终的视频到语言路径并未可靠地利用这些线索。

These results suggest that, in driving, current video-language models remain unreliable temporal reasoners and often behave more like frame-level, bias-driven recognizers than dependable temporal world models.

这些结果表明，在驾驶领域，当前的视频语言模型仍然是不可靠的时间推理工具，其表现往往更像是基于帧、受偏差驱动的识别器，而非可靠的时间世界模型。