LightSTAR framework offers efficient visual document retrieval

Researchers have developed LightSTAR, a new framework for efficient visual document retrieval. This system addresses the computational cost of current methods, which often rely on intensive Multi-modal Large Language Models (MLLMs). LightSTAR employs an LLM-free visual selection stage to quickly narrow down relevant pages using content-grounded query encoding and LLM-free visual embeddings. A subsequent vision-adaptive semantic refinement stage then performs fine-grained matching on these selected candidates, combining textual and layout cues for improved accuracy. Experiments show LightSTAR significantly reduces latency while maintaining state-of-the-art retrieval performance. AI

IMPACT Offers a more efficient alternative to LLM-based methods for visual document retrieval, potentially speeding up research and information access.

RANK_REASON This is a research paper detailing a new framework for visual document retrieval. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• LightSTAR
• Litmaps
• LLM-free visual embeddings
• LLM-free Visual Selection
• Multi-modal Large Language Models
• ScienceCast
• scite Smart Citations
• Vision-adaptive Semantic Refinement

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →