VLMs fail to detect image swaps during self-reflective reasoning with accuracy drops up to 60%, revealing that self-generated reflections do not trigger genuine visual re-examination.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
No citing papers match the current filters.