Text this: An Objective-Dependent and Reproducible Framework for AI Dataset Quality Assessment: In-Domain and Cross-Dataset Evidence From Building-Footprint Segmentation