Starting from February 2025, an updated version of this work can be found at MaskPlanner. The new version includes (1) more training data, (2) a better training pipeline (updated loss function, ...), ...