CVPR 2025 Eval-FoMo 2 workshop: Hidden in plain sight: VLMs overlook their visual representations