Fooling the LVLM Judges: Visual Biases in LVLM-Based Evaluation