Interactive Segmentation
Interactive segmentation aims to improve the efficiency and accuracy of image segmentation by incorporating user input, such as clicks or scribbles, to guide the segmentation process. Current research focuses on enhancing the efficiency and accuracy of existing models like Segment Anything Model (SAM), developing novel architectures that leverage transformers and graph neural networks, and addressing challenges like handling diverse prompt types, scale variations, and uncertainty in perception, particularly in 3D and medical imaging contexts. This field is significant because it reduces the need for extensive manual annotation, enabling faster and more accurate segmentation in various applications, including robotics, medical image analysis, and remote sensing.
Papers
Embodied Uncertainty-Aware Object Segmentation
Xiaolin Fang, Leslie Pack Kaelbling, Tomás Lozano-Pérez
SAM 2 in Robotic Surgery: An Empirical Evaluation for Robustness and Generalization in Surgical Video Segmentation
Jieming Yu, An Wang, Wenzhen Dong, Mengya Xu, Mobarakol Islam, Jie Wang, Long Bai, Hongliang Ren