OVFact is a reference-free metric combining precision and recall for long captions, and using it to filter noisy VLM-generated training data improves factual accuracy without sacrificing descriptiveness.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
OVFact: Measuring and Improving Open-Vocabulary Factuality for Long Caption Models
OVFact is a reference-free metric combining precision and recall for long captions, and using it to filter noisy VLM-generated training data improves factual accuracy without sacrificing descriptiveness.