Abstract: Visual object tracking remains challenging under complex motion, long-term occlusion, and distractor interference. The Segment Anything Model 2 (SAM2) provides a strong foundation for visual ...
Abstract: The Segment Anything Model 2 (SAM2) demonstrates remarkable universal segmentation capabilities on natural images. However, its performance on ultrasound images is significantly degraded due ...
A two-stage framework for data-efficient instance segmentation that leverages the knowledge of SAM 2. Support two datasets (used in our paper) RarePlanes and SpaceNet, with the option to easily add ...