VisionTrap shows that GPT-4o, GPT-4.1, Gemini Flash 2.5, and LLaVA tend to answer unanswerable visual questions rather than abstain, especially when given multiple-choice options.
Reassessing evaluation practices in visual ques- tion answering: A case study on out-of-distribution general- ization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
VisionTrap: Unanswerable Questions On Visual Data
VisionTrap shows that GPT-4o, GPT-4.1, Gemini Flash 2.5, and LLaVA tend to answer unanswerable visual questions rather than abstain, especially when given multiple-choice options.