For the synthetic-to-real generalization experiments on UAVid and AeroScapes, were models trained on the full SkyScenes dataset, or were they trained on specific (height, pitch) subsets that are aligned with the real datasets, while still leveraging the full diversity of weather conditions and map layouts within those subsets
For the synthetic-to-real generalization experiments on UAVid and AeroScapes, were models trained on the full SkyScenes dataset, or were they trained on specific (height, pitch) subsets that are aligned with the real datasets, while still leveraging the full diversity of weather conditions and map layouts within those subsets