SeC: Advancing Complex Video Object Segmentation via Progressive Concept Construction