By Javed Iqbal and Mohsen Ali

• 2020.12.09: code released for LT-WAN and OS-WAN

0. Introduction
1. Requirements
2. Setup
3. Usage
4. Results
5. Note
6. Citation

This repository contains the weakly supervised learning framwork for domain adaptation of built-up regions segmnentation based on the work described in ISPRS Photogrametery and Remote Sensing 2020 paper "[WAN: Weakly-Supervised Domain Adaptation for Built-up Region Segmentation in Aerial and Satellite Imagery]". (https://arxiv.org/pdf/2007.02277.pdf).

The code is tested in Ubuntu 16.04. It is implemented based on Keras with tensorflow backend and Python 3.5. For GPU usage, the maximum GPU memory consumption is about 7 GB in a single GTX 1080.

We assume you are working in wan-master folder.

0. Datasets:

• Download Rwanda dataset.
• Put downloaded data in "datasets" folder.

0. Set the PYTHONPATH environment variable:

cd wan-master

1. Adaptation

• OSA: Output space Adaptation:

python adapt_OSA.py --data-dir path_to_dataset_folder --data-list-train training_images_list --data-list-val validation_images_list

• LTA: Output space Adaptation:

python adapt_LTA.py --data-dir path_to_dataset_folder --data-list-train training_images_list --data-list-val validation_images_list

• To run the code, you need to set the data paths of source data (data-root) and target data (data-root-tgt) by yourself. Besides that, you can keep other argument setting as default.

4. Evaluation

5. Train in source domain

python train.py --data-dir path_to_dataset_folder --data-list-train training_images_list --data-list-val validation_images_list

If you found this useful, please cite our paper.

@inproceedings{iqbal2020weakly,
      title={Weakly-supervised domain adaptation for built-up region segmentation in aerial and satellite imagery},
      author={Iqbal, Javed and Ali, Mohsen},
      journal={ISPRS Journal of Photogrammetry and Remote Sensing},       volume={167},       pages={263--275},       year={2020},       publisher={Elsevier} }

Contact: javed.iqbal@itu.edu.pk