Event-STVG: Reasoning-Based Spatio-Temporal Video Grounding
October 14, 2026
Event-STVG shifts video grounding from instance-specific localization to event-level reasoning by integrating long-form context and multi-target grounding. The benchmark includes 4,016 annotated event occurrences across eight different domains.
HOW THIS AFFECTS YOU
●
researcherYou can now benchmark MLLMs on their ability to reason about complex, repeated events in video.