Training-Free From survey PDF — verify links
QVAD
QVAD — Agentic Pipelines
arXiv · 2026Human-reviewed summary status
Survey summary
Treats VLM-LLM interaction as a dynamic dialogue instead of static prompting: an LLM agent iteratively refines queries from visual context so lightweight VLMs produce high-fidelity captions, reaching state-of-the-art results with a fraction of the parameters of larger training-free methods.
Taxonomy placement
Training-Free
Training-FreeAgentic Pipelines
Method characteristics
Training-FreeQuestion-centricAgenticEfficient