A multi-stage multimodal RAG pipeline with intent refinement, API selection, and relevance/safety filtering reports higher win rates than baseline MLLMs and near-oracle accuracy on Encyclopedic VQA.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
CUE-M: Contextual Understanding and Enhanced Search with Multimodal Large Language Model
A multi-stage multimodal RAG pipeline with intent refinement, API selection, and relevance/safety filtering reports higher win rates than baseline MLLMs and near-oracle accuracy on Encyclopedic VQA.