Seeing but Not Believing: Probing the Disconnect Between Visual Attention and Answer Correctness in VLMs

Open in new window