Informative Path Planning Framework for Active Learning in UAV-based Semantic Mapping

This dataset contains semantic segmentation train-validation-test data splits of (1) the ISPRS Potsdam orthomosaic (https://www.isprs.org/education/benchmarks/UrbanSemLab/2d-sem-label-potsdam.aspx), (2) the RIT-18 landcover orthomosaic (https://github.com/rmkemker/RIT-18), and (3) the industrial environment of the photorealistic Flightmare quadrotor simulator (https://github.com/uzh-rpg/flightmare). All splits are generated by simulating UAV missions at fixed altitudes. We use these datasets in our "An Informative Path Planning Framework for Active Learning in UAV-based Semantic Mapping" paper. Further, it contains (4) model checkpoints of our proposed Bayesian ERFNet framework (https://github.com/dmar-bonn/bayesian_erfnet) pre-trained on Cityscapes, (5) the ISPRS Potsdam and RIT-18 RGB and semantically labelled orthomosaics, and (6) the Flightmare render binary for the industrial environment.

UAV Zurich Semantic Segmentation of Crops and Weeds

Images captured by an UAV at Zurich with additional annotations of crops and weeds (2017). The *split.yaml* file contains information about which images belong to the train, val, and test set.

UAV Bonn Semantic Segmentation of Crops and Weeds

Images captured by an UAV at Bonn with additional annotations of crops and weeds (2017). The *split.yaml* file contains information about which images belong to the train, val, and test set.