📄 Paper | 🌐 Project Website | 🤗 HuggingFace | 🗄️ Zenodo | 🏆 LeaderBoard | 🤖 Baseline Models

Mirali Purohit^1,3, Bimal Gajera^1*, Vatsal Malaviya^1*, Irish Mehta^1*,
Kunal Kasodekar¹, Jacob Adler², Steven Lu³, Umaa Rebbapragada³, Hannah Kerner¹

¹School of Computing and Augmented Intelligence, Arizona State University, Tempe, AZ, USA
²School of Earth and Space Exploration, Arizona State University, Tempe, AZ, USA
³Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA
^*Equal Contribution

---

Introduction

Mars-Bench introduces the first standardized benchmark for Mars science, covering 20 datasets across classification, segmentation, and object-detection tasks using both orbital and rover imagery. It includes a wide range of geologic features; such as craters, cones, boulders, and frost to support comprehensive evaluation.

Representative samples from selected Mars-Bench datasets from all three task categories.

Overview of Mars-Bench datasets across all three task categories. To distinguish the benchmarked versions from their original sources, all dataset names are prefixed with 'mb-', which indicates Mars-Bench. Observation sources are labeled as O (Orbiter) and R (Rover).

---

Installation

# Install the package with core dependencies
pip install -e .

# Install with development dependencies (for testing, linting, etc.)
pip install -e ".[dev]"

Usage

Mars-Bench uses a Hydra-based CLI with marsbench.main as the entry point:

• Quickstart examples: Refer to EXAMPLES.md for end-to-end commands for classification, segmentation, and detection, including how to select datasets and models.
• Configs and benchmarking scripts: Explore the marsbench/configs/ directory for all task/model/data configurations and the benchmark/ folder for scripts used to reproduce the paper’s experiments and figures.

Citation

If you use Mars-Bench in your research, please use the following citation:

@inproceedings{purohit2025marsbench,
    title={Mars-Bench: A Benchmark for Evaluating Foundation Models for Mars Science Tasks},
    author={Mirali Purohit and Bimal Gajera and Vatsal Malaviya and Irish Mehta and Kunal Sunil Kasodekar and Jacob Adler and Steven Lu and Umaa Rebbapragada and Hannah Kerner},
    booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track},
    year={2025},
    url={https://arxiv.org/pdf/2510.24010}
}

Contact Information

Please reach out to Mirali Purohit mpurohi3@asu.edu, if you have any queries or issues regarding Mars-Bench.