A dataset with high resolution (4K) images and manually-annotated dense labels every 50 frames.
Source: Semantics through Time: Semi-supervised Segmentation of Aerial Videos with Iterative Label Propagation