Stage-wise annotations on CRAG-MM show that knowledge retrieval and reasoning, not visual grounding or object identification, is the dominant bottleneck in knowledge-intensive VQA.
Title resolution pending
1 Pith paper cite this work, alongside 1 external citations. Polarity classification is still indexing.
1
Pith paper citing it
1
external citations · OpenAlex
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
CRAG-MM-Diagnostics: Enabling Stage-Wise Analysis of Knowledge-Intensive VQA
Stage-wise annotations on CRAG-MM show that knowledge retrieval and reasoning, not visual grounding or object identification, is the dominant bottleneck in knowledge-intensive VQA.