@inproceedings{liang-etal-2025-investigating,
    title = "Investigating and Enhancing the Robustness of Large Multimodal Models Against Temporal Inconsistency",
    author = "Liang, Jiafeng  and
      Jiang, Shixin  and
      Dong, Xuan  and
      Wang, Ning  and
      Chu, Zheng  and
      Su, Hui  and
      Fu, Jinlan  and
      Liu, Ming  and
      Ng, See-Kiong  and
      Qin, Bing",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingestion-acl-25/2025.acl-long.692/",
    pages = "14149--14162",
    ISBN = "979-8-89176-251-0",
    abstract = "Large Multimodal Models (LMMs) have recently demonstrated impressive performance on general video comprehension benchmarks. Nevertheless, for broader applications, the robustness of their temporal analysis capability needs to be thoroughly investigated yet predominantly ignored. Motivated by this, we propose a novel temporal robustness benchmark (TemRobBench), which introduces temporal inconsistency perturbations separately at the visual and textual modalities to assess the robustness of models. We evaluate 16 mainstream LMMs and find that they exhibit over-reliance on prior knowledge and textual context in adversarial environments, while ignoring the actual temporal dynamics in the video. To mitigate this issue, we design panoramic direct preference optimization (PanoDPO), which encourages LMMs to incorporate both visual and linguistic feature preferences simultaneously. Experimental results show that PanoDPO can effectively enhance the model{'}s robustness and reliability in temporal analysis."
}