ReVisionLLM: Recursive Vision-Language Model for Temporal Grounding in Hour-Long Videos
This addresses the challenge of accurately localizing events in extended video content for applications like video analysis and retrieval, representing a novel advancement rather than an incremental improvement.
The paper tackled the problem of temporal grounding in hour-long videos, where existing vision-language models struggle due to frame limitations, and proposed ReVisionLLM, a recursive model that outperformed previous state-of-the-art methods by +2.6% R1@0.1 on the MAD dataset.
Large language models (LLMs) excel at retrieving information from lengthy text, but their vision-language counterparts (VLMs) face difficulties with hour-long videos, especially for temporal grounding. Specifically, these VLMs are constrained by frame limitations, often losing essential temporal details needed for accurate event localization in extended video content. We propose ReVisionLLM, a recursive vision-language model designed to locate events in hour-long videos. Inspired by human search strategies, our model initially targets broad segments of interest, progressively revising its focus to pinpoint exact temporal boundaries. Our model can seamlessly handle videos of vastly different lengths, from minutes to hours. We also introduce a hierarchical training strategy that starts with short clips to capture distinct events and progressively extends to longer videos. To our knowledge, ReVisionLLM is the first VLM capable of temporal grounding in hour-long videos, outperforming previous state-of-the-art methods across multiple datasets by a significant margin (+2.6% R1@0.1 on MAD). The code is available at https://github.com/Tanveer81/ReVisionLLM.