Published August 2, 2026 | Version v1

AdaSemSeg: Processed Seismic Facies Datasets (F3, Parihaka, Penobscot)

  • 1. ROR icon University of Utah
  • 2. ROR icon Microsoft (United States)
  • 3. The University of Utah, Salt Lake City, UT, USA

Description

Processed seismic facies datasets accompanying the paper "AdaSemSeg: An Adaptive Few-Shot Semantic Segmentation of Seismic Facies" (Saha & Whitaker, IEEE
Transactions on Geoscience and Remote Sensing, 2025, DOI: 10.1109/TGRS.2025.3595010).

Contents:
- F3_data.zip — F3 (Netherlands) seismic volume and facies labels, using the train/val/test split from the original F3 benchmark release.
- Parihaka_data.zip — Parihaka (New Zealand) seismic volume and facies labels from the SEG/NZP&M Parihaka-3D open dataset.
- Penobscot_data.zip — Penobscot (Canada) seismic volume and facies labels, processed from the original Baroni et al. release (arXiv:1903.12060) by
  cropping to remove corrupted/noisy volume edges and rescaling via 5-95 percentile clipping. The exact processing script (data/Penobscot/process_data.py)
  is included in the companion GitHub repository for full transparency and reproducibility.

Each archive contains the seismic and label volumes as .npy arrays plus the corresponding train/val/test split JSON used in the paper's experiments.

Companion code, trained model checkpoints (Zenodo DOI: 10.5281/zenodo.21762769), and usage instructions are at: https://github.com/Surojit-Utah/AdaSemSeg.

Files

F3_data.zip

Files (2.6 GB)

Name Size
md5:c6965cd27f9dc829c5afefbd06c891f3
1.1 GB Preview Download
md5:10182656d7d1e5400cb20c3d99cf78f2
449.9 MB Preview Download
md5:351b85fe82f28e1bae24d39ca5fafdae
1.1 GB Preview Download

Additional details

Software

Repository URL
https://github.com/Surojit-Utah/AdaSemSeg
Programming language
Python