Temporal-Oriented Recipe for Transferring Large Vision-Language Model to Video Understanding
This work addresses the challenge of adapting LVLMs for video understanding, which is crucial for applications in video analysis, but it is incremental as it builds on existing LVLM frameworks.
The paper tackled the problem of limited temporal understanding in large vision-language models (LVLMs) for video tasks by identifying key components at the visual encoder-LLM interface and proposing a temporal-oriented recipe, resulting in significant performance improvements on standard video understanding benchmarks.
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.