What is AIRS?
AIRS (Aerial Imagery for Roof Segmentation) is a public dataset that aims at benchmarking the algorithms of roof segmentation from very-high-resolution aerial imagery. The main features of AIRS can be summarized as:
- 457km2 coverage of orthorectified aerial images with over 220,000 buildings
- Very high spatial resolution of imagery (0.075m)
- Refined ground truths that strictly align with roof outlines
AIRS dataset covers almost the full area of Christchurch, the largest city in the South Island of New Zealand. The photography was taken during the flying seasons of 2015 and 2016, and the supplied images are ortho-rectified DOMs with RGB channels and 7.5cm resolution in projection of New Zealand Transverse Mercator. There are 226,342 labeled buildings within the whole area for experiment. To eliminate the impact of relief displacement, the ground truths for buildings are carefully refined to align with their roofs. Therefore, the segmentation task posed for AIRS contains two semantic classes: roof and non-roof pixels. (To know more details about problems introduced by relief displacement, please refer to our paper.)
The AIRS dataset is now fully open and you can freely download the dataset from here.
We strongly encourage you to participate in the AIRS Challenge by submitting your results here. Your results will be evaluated automatically and an accuracy report will be provided. If there is no problem, your results will be shown and ranked in the leaderboard.
If you use our dataset, please cite our paper.
AIRS dataset is created through a careful refinement process on the open-source data released by Land Information of New Zealand, we want to thank National Topographic Office of New Zealand for their kind open-souring of the data.