SegAgent: Exploring Pixel Understanding Capabilities in MLLMs by Imitating Human Annotator Trajectories

Benchmark Model Rank Results
referring-expression-segmentation-on-refcocog-valSegAgent#7Overall IoU: 75.11
referring-expression-segmentation-on-refcoco-plus-valSegAgent#12Overall IoU: 72.49
referring-expression-segmentation-on-refcoco-valSegAgent#16Overall IoU: 79.7