Mixed Signals: Decoding VLMs' Reasoning and Underlying Bias in Vision-Language Conflict

Pouya Pezeshkpour, Moin Aminnaseri, Estevam Hruschka

Vision-language models (VLMs) have demonstrated impressive performance by effectively integrating visual and textual information to solve complex tasks. However, it is not clear ho…