Temporal-Oriented Recipe for Transferring Large Vision-Language Model to Video Understanding

Thong Nguyen1, Zhiyuan Hu1, Xu Lin1, Cong-Duy Nguyen2, See-Kiong Ng1, Luu Anh Tuan2,
1National University of Singapore, 2Nanyang Technological University
Paper arXiv Code Twitter (X)

Abstract

Recent years have witnessed outstanding advances of large vision-language models (LVLMs). In order to tackle video understanding, most of them depend upon their implicit temporal understanding capacity. As such, they have not deciphered important components that contribute to temporal understanding ability, which might limit the potential of these LVLMs for video understanding. In this work, we conduct a thorough empirical study to demystify crucial components that influence the temporal understanding of LVLMs. Our empirical study reveals that significant impacts are centered around the intermediate interface between the visual encoder and the large language model. Building on these insights, we propose a temporal-oriented recipe that encompasses temporal-oriented training schemes and an upscaled interface. Our final model developed using our recipe significantly enhances previous LVLMs on standard video understanding tasks.

Temporal-Oriented Recipe

Illustration of Temporal-Oriented Recipe to Transfer Large Vision-Language Model to Video Understanding.

Key Results

Performance tables showing consistent gains from temporal-oriented recipe.
Table 1: Temporal-oriented recipe boosts video understanding performance at both 7B and 13B scales.

BibTeX

@article{nguyen2025temporal,
  title={Temporal-Oriented Recipe for Transferring Large Vision-Language Model to Video Understanding},
  author={Nguyen, Thong and Hu, Zhiyuan and Lin, Xu and Nguyen, Cong-Duy and Ng, See-Kiong and Tuan, Luu Anh},
  journal={arXiv preprint arXiv:2505.12605},
  year={2025}
}