Designing Efficient Multimodal Systems With Vision-Centric Inputs. IJCHML 2025, 3 (3).