Enhancing Multimodal Models With Vision-Centric Feedback Loops. IJCHML 2025, 3 (1).