VideoLoom: A Video Large Language Model for Joint Spatial-Temporal Understanding

Recent advancements in Video Large Language Models (Video LLMs) have demonstrated impressive results, yet existing approaches handle either temporal or spatial dimension in isolation, struggling in the analysis of complex events that require spatial-temporal integration. To bridge this gap, we propose VideoLoom, a unified Video LLM for joint spatial-temporal understanding. To facilitate the development of fine-grained spatial and temporal localization capabilities, we curate LoomData-8.7k, a character-centric video dataset with temporally grounded and spatially localized captions. With this, VideoLoom achieves the state-of-the-art performance across a variety of spatial and temporal benchmarks. In addition, we introduce LoomBench, a benchmark consisting of temporal, spatial, and compositional video–question pairs, with a novel metric $J$&$F_{bi-fore}$, enabling a comprehensive evaluation of Video LLMs from diverse aspects. Collectively, these contributions offer a universal and effective suite for joint spatial-temporal video understanding, setting a new standard in multimodal intelligence.

Understanding a video is more than recognizing objects or actions. To truly understand what is happening, an AI system must know both when an event happens and where it occurs on the screen. Humans do this naturally, but existing methods process time and space separately, making them struggle in complex real-world scenes. In this work, we introduce VideoLoom, a Video LLM designed to connect spatial and temporal understanding within a single framework. We also build a new dataset that teaches the model to jointly track events across both time and space, as well as a new benchmark for evaluating this capability in a comprehensive way. Our approach outperforms previous methods on video tasks that require locating actions and objects in time, in space, or jointly across both. By helping AI systems interpret videos in a more unified and human-like way, our work may support future applications such as video assistants, robotics, and intelligent surveillance systems.