UAV-based imagery for road damage detection

This dataset contains a set of annotated examples of damage present to the road,  obtained via an UAV in Poznań, Poland. The imagery was obtained by flying a UAV over a set of roads at approximately 70m above ground level, and further creating an ortophotomosaics of three different pieces of road. Lastly, the ortophotomosaics were patched into non-overlapping patches of 512x512 pixel size, and each patch was manually labeled with a set of objects and classes proposed in the UAPD dataset (https://doi.org/10.1016/j.autcon.2021.103991, https://github.com/tantantetetao/UAPD-Pavement-Distress-Dataset).

Patches without a significant presence of the road-like objects were discarded, and the data includes:

• ortophotomosaics -- a set of three original ortophotomosaics with ground sampling distance of 2.54 cm/pixel in .tif format,
• labeled patches -- a set of 99 patches containing road with annotations in both PASCAL VOC (.xml) and YOLO (.txt) formats

Lastly, this data has been used as a holdout set in a Master's thesis conducted at Poznań University of Technology.