The authors address knowledge-intensive visual question answering (VQA) where external textual evidence beyond the image is needed.
Confidence-guided complementary evidence pool improves multimodal retrieval-augmented generation
The authors present CLIMB, a training-free framework that constructs a compact, non-redundant evidence set and only updates answers when estimated confidence increases, outperforming standard Top-K retrieval on knowledge-intensive VQA benchmarks.
Big Tech
Hang Gao · Wujiang Xu · Zhixing Zhang · Kai Mei · Jingyi Yang · Dimitris N. Metaxas
Rutgers University · Google
Research Digest··3 min read
The authors propose CLIMB, a training-free inference-time framework for multimodal retrieval-augmented generation.
Why this paper
From Google and Rutgers University
In one line
CLIMB selects complementary evidence and uses confidence scores to decide when to update answers, improving multimodal RAG without retraining.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§