Building footprint detection from satellite/aerial images with Mask R-CNN and image matting (KNN matting, closed-form matting and Grabcut).
Run python detect.py for detection, python evaluate.py for evaluation, and python regularize.py for regularizing the footprint. You may have to change the hardcoded paths in these scripts.
You can download the cropped aerial image tiles and raster labels from WHU Building Dataset.
│ detect.py
│ evaluate.py
│ mrcnn_demo.ipynb
│
├─matting
│ closed_form_matting.py
│ grabcut.py
│ knn_matting.py
│ solve_foreground_background.py
│
├─models
│ models_here.txt
│
├─mrcnn
│ config.py
│ model.py
│ parallel_model.py
│ utils.py
│ visualize.py
│ __init__.py
│
└─polygon
regularize.py