POPEN: Preference-Based Optimization and Ensemble for LVLM-Based Reasoning Segmentation | Read Paper on Bytez