arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.26802cs.CVcs.AIcs.CL
收藏

视频模型准备好作为零样本推理器了吗?基于 MME-CoF 基准的实证研究

Are Video Models Ready as Zero-Shot Reasoners? An Empirical Study with the MME-CoF Benchmark

CUHK(香港中文大学);IMIXR;MMLab;Peking University(北京大学);Northeastern University(东北大学)

Ziyu Guo, Xinyan Chen, Renrui Zhang, Ruichuan An, Yu Qi, Dongzhi Jiang, Xiangtai Li, Manyuan Zhang, Hongsheng Li, Pheng-Ann Heng

AI总结:

本研究基于 MME-CoF 基准对 Veo-3 视频模型进行实证评估,发现其在短程空间和局部动态推理上表现优异,但在长程因果和抽象逻辑上受限,尚不能作为可靠的零样本推理器,但可作为专用推理模型的补充视觉引擎。

摘要:

近期的视频生成模型能够生成高保真、时间连贯的视频,这表明它们可能编码了丰富的世界知识。除了逼真的合成,它们还展现出视觉感知、建模和操纵的涌现行为。然而,一个重要问题依然存在:在充满挑战的视觉推理场景中,视频模型准备好作为零样本推理器了吗?在本研究中,我们开展实证研究以全面探究这一问题,重点关注领先且流行的 Veo-3。我们在 12 个维度上评估其推理行为,包括空间、几何、物理、时间和具身逻辑,系统地刻画其优势与失败模式。为使本研究标准化,我们将评估数据整理为 MME-CoF,这是一个紧凑的基准,能够对 Chain-of-Frame (CoF) 推理进行深入且彻底的评估。我们的发现表明,尽管当前视频模型在短程空间一致性、细粒度定位和局部一致动态上展现出有前景的推理模式,但在长程因果推理、严格几何约束和抽象逻辑方面仍显局限。总体而言,它们作为独立的零样本推理器尚不可靠,但作为专用推理模型的补充视觉引擎,展现出了令人鼓舞的迹象。项目页面:https://video-cof.github.io

英文摘要:

Recent video generation models can produce high-fidelity, temporally coherent videos, indicating that they may encode substantial world knowledge. Beyond realistic synthesis, they also exhibit emerging behaviors indicative of visual perception, modeling, and manipulation. Yet, an important question still remains: Are video models ready to serve as zero-shot reasoners in challenging visual reasoning scenarios? In this work, we conduct an empirical study to comprehensively investigate this question, focusing on the leading and popular Veo-3. We evaluate its reasoning behavior across 12 dimensions, including spatial, geometric, physical, temporal, and embodied logic, systematically characterizing both its strengths and failure modes. To standardize this study, we curate the evaluation data into MME-CoF, a compact benchmark that enables in-depth and thorough assessment of Chain-of-Frame (CoF) reasoning. Our findings reveal that while current video models demonstrate promising reasoning patterns on short-horizon spatial coherence, fine-grained grounding, and locally consistent dynamics, they remain limited in long-horizon causal reasoning, strict geometric constraints, and abstract logic. Overall, they are not yet reliable as standalone zero-shot reasoners, but exhibit encouraging signs as complementary visual engines alongside dedicated reasoning models. Project page: https://video-cof.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。

↑