Presentation
This repository contains the code of DISCA from our paper: Interactive Learning for Semantic Segmentation in Earth Observation. In a nutshell, it consists in neural networks trained to perform semantic segmentation with human guidance. This builds on our previous work DISIR.
This repository is divided into two parts:
train
which contains the training code of the networks (README)qgs_plugin
which contains the code of the QGIS plugin used to perform the interactive segmentation (README)
Install Python dependencies
conda create -n disca python=3.7 rtree gdal=2.4 opencv scipy shapely -c 'conda-forge'
conda activate disca
pip install -r requirements.txt
To use
Please note that this repository has been tested on Ubuntu 18.4, QGIS 3.8 and python 3.7 only.
- Download a segmentation dataset such as ISPRS Potsdam or INRIA dataset.
- Prepare this dataset according to
Dataset preprocessing
intrain/README.md
. - Train a modelstill following
train/README.md
. - Install the QGIS plugin following
qgs_plugin/README.md
. - Follow
How to start
inqgs_plugin/README.md
and start segmenting your data !
References
If you use this work for your projects, please take the time to cite our ECML-PKDD MACLEAN Workshop paper:
@inproceedings{lenczner2020interactive,
author = {Lenczner, G. and Chan-Hon-Tong, A. and Luminari, N. and Le Saux, B. and Le Besnerais, G.},
title = {Interactive Learning for Semantic Segmentation in Earth Observation},
booktitle = {ECML-PKDD MACLEAN Workshop},
year = {2020}
}
Licence
Code is released under the MIT license for non-commercial and research purposes only. For commercial purposes, please contact the authors.
See LICENSE for more details.
Authors
See AUTHORS.md
Acknowledgements
This work has been jointly conducted at Delair and ONERA-DTIS.