[Submitted on 13 May 2026 (v1), last revised 29 Oct 2026 (this version, v3)]
Abstract:Video temporal grounding (VTG) takes an untrimmed video and a natural-language query as input and localizes the temporal moment that best matches the query. Existing methods rely on large, task-specific datasets requiring costly manual annotation. We introduce EvoGround, a framework of two coupled self-evolving agents, a proposer and a solver, that learn temporal grounding from raw videos without any task-specific annotations. The proposer generates query–moment pairs from raw videos, while the solver learns to ground them and feeds back signals that improve the proposer in return. Through this self-reinforcing reinforcement-learning loop, the two agents are initialized from the same backbone and mutually improve across iterations. Trained on 2.5K unlabeled videos, EvoGround matches or surpasses most fully supervised models across five VTG benchmarks, while emerging as a state-of-the-art fine-grained video captioner. We publicly release all resources at this https URL.
Submission history
From: Minjoon Jung [view email]
[v1] Wed, 13 May 2026 17:25:51 UTC (10,690 KB)
[v2] Fri, 23 Oct 2026 18:46:22 UTC (10,754 KB)
[v3] Thu, 29 Oct 2026 15:06:05 UTC (10,754 KB)
