SegAgent: Exploring Pixel Understanding Capabilities in MLLMs by Imitating Human Annotator Trajectories Paper • 2503.08625 • Published Mar 11, 2025 • 27