Awesome

Deep Learning-based Building Footprint Extraction with Missing Annotations

Jian Kang, Ruben Fernandez-Beltran, Xian Sun, Jingen Ni, Antonio Plaza

This repo contains the codes for the GRSL paper: Deep Learning-based Building Footprint Extraction with Missing Annotations. We propose a novel loss function for extracting building footprints based on the training images with missing annotations. The loss includes three terms: 1) logit adjusted cross entropy (LACE) loss, aimed at discriminating between building and background pixels from a long-tailed label distribution; 2) weighted dice loss, aimed at increasing the F1 scores of the predicted building masks; and 3) boundary alignment loss, which is optimized for preserving the fine-grained structure of building boundaries.

Usage

train/main_.py is the script of the proposed method for training and validation.

utils/loss_.py contains the considered losses for building footprint extraction.

Some codes are modified from SegLoss and Geoseg.

Citation

@article{kang2021bdsegma,
  title={{Deep Learning-based Building Footprint Extraction with Missing Annotations}},
  author={Kang, Jian and Fernandez-Beltran, Ruben and Sun, Xian and Ni, Jingen and Plaza, Antonio},
  journal={IEEE Geoscience and Remote Sensing Letters},
  year={2021},
  note={DOI:10.1109/LGRS.2021.3072589}
  publisher={IEEE}
}