Apple ML Research introduces GH-ESD for error slice discovery in vision tasks

Apple ML Research proposes GH-ESD, a grounded hypothesis-driven approach to discover systematic error slices in instance-level vision tasks, aiming to improve model robustness and evaluation.