nasa-ibm-ai4science/Sombench-IMP-Segmentation

SomBench Benchmark: Irregular Mare Patch (IMP) Segmentation

Science theme: Volcanic history

Task: Binary semantic segmentation

Dataset Summary

A binary semantic-segmentation benchmark for irregular mare patches (IMPs): rare, morphologically subtle features interpreted as unusually young volcanic landforms. Each sample is an LROC NAC image tile paired with a binary IMP mask (IMP vs. background). The set is derived from published IMP polygon annotations, framed as a coarse-to-fine mapping problem, and split into train / validation / test.

Dataset Structure

Directory layout

IMP_dataset/
├── README.md                  # dataset notes (source)
├── all/                       # image + mask tiles
└── *.txt                      # split file lists

Each sample is a pair of GeoTIFFs sharing a common stem:

{PRODUCT}.ech.cog__target_{id}__idx_{id}_p0_img.tif        # NAC image tile
{PRODUCT}.ech.cog__target_{id}__idx_{id}_p0_mask.tif       # binary IMP mask (used for training/eval)

• {PRODUCT}: source LROC NAC EDR id (e.g. M1119107675LE), calibrated, map-projected and stored as a Cloud-Optimized GeoTIFF (.ech.cog).
• {id}: id for the sample.
• _mask.tif: the label used for training/evaluation.

Contents

Item	Value
Patch size	256 × 256 px (256 m; NAC ≈ 1 m/pixel)
Source imagery	LROC NAC (map-projected COG)
Files	260 GeoTIFFs ≈ 130 img / mask pairs
Splits	train/, val/, test/ (membership also listed in the .txt split files)
Per-split counts	Train 100 · Val 20 · Test 10 (130 tiles total)
Classes	binary: IMP (foreground) vs. background

Known Limitations

• Noisy labels: instrument pointing/geolocation uncertainty can misalign IMP annotations by tens of meters; a subset was manually re-aligned, but residual offsets remain.
• IMP polygons exceeding the 256 m patch side length (≈ 100 features) were excluded.

Citation

@article{fraccaro2026lfm,
  title  = {Multimodal-Multiresolution Foundation Model for Lunar Remote Sensing},
  author = {Fraccaro, Paolo and Nyirjesy, Gabby and Szwarcman, Daniela and Patil, Himanshu
            and Gaur, Vishal and Lal, Rohit and Slank, Rachel A. and Dawson, Geoffrey
            and Debary, Hiyam and Dionelis, Nikolaos and Barker, Michael K. and Annex, Andrew
            and Viswanathan, Vishnu and Morse, Zachary and Schaefer, Ethan I. and Kumar, Ankur
            and Watson, Campbell D. and Dawson-Rigas, Rebekah I. and Maskey, Manil
            and Roy, Sujit and Ramachandran, Rahul and Bernab\'e-Moreno, Juan},
  year   = {2026}
  howpublished = {\url{https://huggingface.co/collections/nasa-ibm-ai4science/nasa-ibm-lunar-fm-and-downstream-models}}
}

@misc{sombench2026collection,
  author = {Patil, Himanshu and Nyirjesy, Gabby and Slank, Rachel A. and Gaur, Vishal
          and Szwarcman, Daniela and Fraccaro, Paolo and Dionelis, Nikolaos and Barker, Michael K.
          and Annex, Andrew and Viswanathan, Vishnu and Morse, Zachary and Schaefer, Ethan I.
          and Debary, Hiyam and Kumar, Ankur and Lal, Rohit and Dawson, Geoffrey
          and Watson, Campbell and Dawson-Rigas, Rebekah I. and Maskey, Manil
          and Bernab\'e-Moreno, Juan and Ramachandran, Rahul and Roy, Sujit},
  title        = {{SomBench}: Benchmark Dataset for Advancing Machine Learning in Lunar Science},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/collections/nasa-ibm-ai4science/lunar-fm-ml-ready-benchmark-dataset-sombench}}
}

License

Released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.