fyaft2012 / geomars

Code for "DoMars16k: A Diverse Dataset for Weakly Supervised Geomorphologic Analysis on Mars"

DoMars16k: A Diverse Dataset for Weakly Supervised Geomorphologic Analysis on Mars

Paper | Data

Thorsten Wilhelm, Melina Geis*, Jens Püttschneider*, Timo Sievernich*, Tobias Weber*, Kay Wohlfarth and Christian Wöhler, TU Dortmund University, * These authors contributed equally to this work.

This repository contains the code to recreate the tables and figures of the paper "DoMars16k: A Diverse Dataset for Weakly Supervised Geomorphologic Analysis on Mars"

Results

Automated Geomorphic Map Creation

Classification Metrics

Classes

Usage

Installation

Create a conda environment from the requirements

conda create --name geomars --file requirements.txt

Run setup.py to download the pre-trained networks and the dataset. The code was tested with Ubuntu 18.04.4 LTS, an NVIDIA RTX6000, and CUDA 10.1.

Configuration

The individual scripts can be adjusted by modifying the corresponding entries in the hyper_params dictionary. For evaluating different models or creating maps with different models only the model entry needs to be changed.

hyper_params = {
    'batch_size': 64,
    'num_epochs': 30,
    'learning_rate': 1e-2,
    'optimizer': 'sgd',
    'momentum': 0.9,
    'model': 'densenet161',
    'num_classes': 15,
    'pretrained': True,
    'transfer_learning': False,
}

Training

Run training.py to train a neural network according to the definitions in the hyper_params dictionary.

Evaluation

Run eval.py to load a pre-trained network from the models folder according to the definitions in the hyper_params dictionary.

Mapping

Run make_map.py to load a pre-trained network from the models folder according to the definitions in the hyper_params dictionary. The analysed image can be changed by modifying CTX_stripe. Three CTX images are already configured:

• Jezero Crater - D14_032794_1989_XN_18N282W
• Oxia Planum - F13_040921_1983_XN_18N024W
• Lycus Sulci - G14_023651_2056_XI_25N148W

Citation

If you find this work useful please consider citing:

Wilhelm, T.; Geis, M.; Püttschneider, J.; Sievernich, T.; Weber, T.; Wohlfarth, K.; Wöhler, C. DoMars16k: A Diverse Dataset for Weakly Supervised Geomorphologic Analysis on Mars. Remote Sens. 2020, 12, 3981.

@article{wilhelm2020domars16k,
  doi = {10.3390/rs12233981},
  url = {https://doi.org/10.3390/rs12233981},
  year = {2020},
  month = dec,
  publisher = {{MDPI} {AG}},
  volume = {12},
  number = {23},
  pages = {3981},
  author = {Thorsten Wilhelm and Melina Geis and Jens P\"{u}ttschneider and Timo Sievernich and Tobias Weber and Kay Wohlfarth and Christian W\"{o}hler},
  title = {{DoMars}16k: A Diverse Dataset for Weakly Supervised Geomorphologic Analysis on Mars},
  journal = {Remote Sensing}
}