3VL: Using Trees to Improve Vision-Language Models' InterpretabilityNir YellinekLeonid Karlinskyet al.2025IEEE TIP