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.
Large scale gan training for high fidelity natural image synthesis,
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
-
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.