This QGIS plugin regularizes building footprints detected with deep learning, including binary semantic and instance segmentation.
The regularization approach projectRegularization was implemented from the publication Machine-learned regularization and polygonization of building segmentation masks", ICPR 2021 on the MapAI: Precision in building segmentation challenge. We have developed an end-to-end binary semantic segmentation workflow in PyTorch and applied it on the MapAI dataset. The developed framework mapAI regularizaton can be used to produce building segmentation maps to test the plugin.
Note that the plugin supports other segmentations too that are in the appropriate image format.
- segmented image either in .png or .tif
- regularized image in .png or .tif
- vector file from regularization in .gpkg
The GUI is presented below and the functionality is shown.
Note that this repo is a work in progress, the expected working version will be published soon.
