SAM-Guided Graph Prompt Reasoning for Cross-Domain Few-Shot Image Segmentation
A network (GPRN) that leverages the Segment Anything Model to improve cross-domain few-shot semantic segmentation. SAM-generated masks are converted into semantic visual prompts, a graph reasoning module (a graph convolutional network over the prompts) enforces global semantic consistency among fragmented prompts, and a non-parametric adaptive point-selection module feeds representative points back to SAM to refine masks at test time.
2501.00303
This computer-vision paper introduces GPRN, a network that uses the Segment Anything Model (SAM) to tackle cross-domain few-shot segmentation (CD-FSS), where a model trained on one domain must segmen…