Awesome

ConeQuest

ConeQuest is the first expert-annotated publicly available dataset for cone segmentation across three different regions on Mars, along with metadata for each sample.

• ConeQuest can be accessed from Zenodo.

• The ConeQuest dataset has a total of 13,686 patches from 8 different subtiles across 3 regions.

• Metadata of each CTX tile across three regions used in the creation of ConeQuest:

More details about ConeQuest are available in this paper.

Getting Started

Environment Setup

conda env create -f conequest_env.yml

Download ConeQuest from Zenodo

./download_data.sh

Training and Testing models

Utilize the arguments described in main.py to train and test models for various configurations. A few important arguments are explained below:

• Provide training_type argument 1 or 2 to train a model for benchmarks 1 and 2, respectively.
• In the training_data_list, provide a list (string separated with a comma) of region/s or size/s on which model will be trained, e.g., "Isidis Planitia, Hypanis" or "small, medium".
• Use if_training_positives to train your model only on positive samples.

python main.py \
    --if_training \
    --training_type 1 \
    --training_data_list "Isidis Planitia" \
    --train_model DeepLab

• In the eval_data, provide which region or size to evaluate.

python main.py \
    --if_testing \
    --training_type 1 \
    --training_data_list "Isidis Planitia, Hypanis" \
    --train_model DeepLab
    --eval_data Hypanis

License

ConeQuest has a Creative Commons Zero v1.0 Universal license.

Citation

If you use ConeQuest in your research, please use the following citation:

@InProceedings{Purohit_2024_WACV,
    author={Purohit, Mirali and Adler, Jacob and Kerner, Hannah},
    title={ConeQuest: A Benchmark for Cone Segmentation on Mars},
    booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month={January},
    year={2024},
    pages={6026-6035}
}