Abstract: Large-scale pre-trained visual foundation models, such as the Segment Anything Model 2 (SAM2), demonstrate strong performance in video segmentation. However, they require multiple iterations ...