arXiv cs.AI· Taekoan Yoo, Wonkyung Jung, Kyunghun Kim, Kyeongbo Kong·· 11 小时前AI 评分34
PhysTrace：视频生成模型物理一致性的可解释时序评估框架
PhysTrace: Interpretable Temporal Evaluation of Physical Consistency in Video Generative Models
AI 导读
PhysTrace 是一个无需训练的框架，利用 MLLM 构建以历史为条件的物理预期，并通过时序视频-语言对齐追踪其与观测视频片段的一致性。在 IntPhys2 上，PhysTrace 实现 67.2% 的可能-不可能判别准确率，超过所评估的直接 MLLM 基线；在新提出的 IntPhys2-Timing 上，违规区间定位 Hit Rate 最高达 77.4%。
正文 · 原文
Abstract:World action models increasingly rely on pretrained diffusion/flow-based video generators assumed to capture physical dynamics for prediction, planning, and action generation. Existing evaluations assess plausibility, probe representations, or measure predictive surprise, yet physical consistency is history-dependent: preceding events constrain what can physically occur later. We introduce PhysTrace, a training-free framework that uses MLLMs to formulate history-conditioned physical expectations and tracks their agreement with observed video windows through temporal video--language alignment, rather than relying on direct plausibility judgments or task-specific probes. On IntPhys2, PhysTrace achieves 67.2% possible--impossible discrimination accuracy, outperforming the evaluated direct MLLM baselines. We further introduce reference-guided masked denoising for controlled reconstruction from mixed possible--impossible video latents. To support these reconstructions and evaluate temporal localization as an additional capability, we introduce IntPhys2-Timing, adding violation intervals to IntPhys2, with PhysTrace achieving a Hit Rate of up to 77.4%. On these reconstructions, human evaluations reveal model-dependent differences in physical consistency and fewer physically impossible outcomes at higher initial noise levels, trends broadly captured by PhysTrace.
| Comments: | 19 pages, 8 figures, 7 tables, including supplementary material |
| Subjects: | Artificial Intelligence (cs.AI); Robotics (cs.RO) |
| Cite as: | arXiv:2610.31429 [cs.AI] |
| (or arXiv:2610.31429v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.31429 arXiv-issued DOI via DataCite |
Submission history
From: Taekoan Yoo [view email]
[v1]
Thu, 22 Oct 2026 15:27:58 UTC (17,536 KB)
[v2]
Thu, 5 Nov 2026 15:42:57 UTC (13,569 KB)
来源：arXiv cs.AI · arxiv.org