SegAgent: Exploring Pixel Understanding Capabilities in MLLMs by Imitating Human Annotator Trajectories Paper • 2503.08625 • Published about 21 hours ago • 22