SegAgent: Exploring Pixel Understanding Capabilities in MLLMs by Imitating Human Annotator Trajectories Paper • 2503.08625 • Published 7 days ago • 24