AI
CoEvoWhen: Policy-Tool Coevolution for Ultra-Long Video Temporal Grounding
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 29/09/2026, 20:00:00 UTC
Briefing created: 01/10/2026, 09:22:40 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 03:18:24 UTC.
Briefings
This article proposes a novel policy-tool coevolution framework for ultra-long video temporal grounding. The method jointly evolves high-level policies and executable media tools from VLM reasoning trajectories, creating reusable skills without model parameter updates. An external skill updater distills transferable experience from long-video grounding tasks to refine orchestration and upgrade tools. Experiments across five benchmarks and three VLMs confirm that this approach consistently improves temporal grounding accuracy while reducing visual token costs during inference.
Key points
- The framework jointly evolves high-level policies and executable media tools from VLM reasoning trajectories to create reusable skills without updating model parameters.
- An external skill updater distills transferable task experience from long-video temporal grounding to refine orchestration and upgrade tools during evolution.
Context and limitations
Results are limited to five benchmarks and three VLMs; performance gains on general long-video QA were observed without additional task-specific evolution.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 09:22:40 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeicfmzo65gjfzzxhsadvi3giomctm4w6ezijo7eiwd6door7p3zrmu