MatchGeo v1.3

A curated multi-region Digital Elevation Model (DEM) dataset for training and benchmarking local feature matching algorithms in urban and natural terrain analysis.

🎯 Overview

MatchGeo aggregates high-resolution elevation data from 13 distinct environments across 6 continents to support research in cross-domain local feature detection and matching. The dataset provides standardised 256x256-pixel patches with handcrafted and automated ground-truth annotations for training computer vision models on geospatial data.

Key Features

• Multi-source fusion: LiDAR, photogrammetry, Structure-from-Motion (SfM), and satellite stereophotogrammetry
• Global coverage: 13 regions across 6 continents — from Antarctica to the Sahara
• Standardised format: All cities processed to 256x256 pixel Cloud Optimized GeoTIFF (COG) tiles
• Rich annotations: 40,287 verified patch pairs (GER_BN and BRA_SP)
• Cross-area evaluation: Explicit intra-region and inter-region test splits
• FAIR compliant: ISO 19115-2 metadata, DOI registration, open access

🗺️ Dataset Coverage

Total Size: ~12.5 GB
Total Tiles: 27,268 (256x256 px patches)
Labelled Tiles: 1,190 (369 Bonn + 821 São Paulo)
Ground Truth Annotations: 40,287 handcrafted point annotations (Bonn + São Paulo)

📊 Per-Region Statistics

📁 Repository Structure

MatchGeo-DEM-v1/
├── README.md                          # This file
├── DATASET_DESCRIPTION.md             # FAIR-compliant formal description
├── LICENSE                            # CC BY 4.0 full legal text
├── CITATION.cff                       # Machine-readable citation
├── manifest.json                      # Central catalog (JSON-LD)
├── checksums.sha256                   # File integrity verification
│
├── data/
│   ├── ATA_MV/
│   │   ├── ATA_MV.tif                 # Merged DEM (BigTIFF, tiled, DEFLATE)
│   │   ├── ATA_MV_extent.geojson      # Bounding polygon
│   │   ├── ATA_MV_tiles.geojson       # Tile index
│   │   ├── ATA_MV_metadata.json       # ISO 19115-2 + OGC 23-008r3 metadata
│   │   ├── ATA_MV.qmd                 # QGIS layer metadata
│   │   └── tiles/                     # 256x256 pixel patches
│   ├── BRA_SP/
│   │   ├── BRA_SP.tif
│   │   ├── BRA_SP_extent.geojson
│   │   ├── BRA_SP_tiles.geojson
│   │   ├── BRA_SP_metadata.json
│   │   ├── BRA_SP.qmd
│   │   ├── annotations/               # JSON keypoint files (labelled)
│   │   │   └── BRA_SP_###_###.json    # Handcrafted annotations
│   │   └── tiles/
│   ├── ... (11 more cities)
│   └── GER_BN/
│       ├── GER_BN.tif
│       ├── GER_BN_extent.geojson
│       ├── GER_BN_tiles.geojson
│       ├── GER_BN_metadata.json
│       ├── GER_BN.qmd
│       ├── annotations/               # Handcrafted annotations
│       │   └── GER_BN_###_###.json
│       └── tiles/
│
├── splits/
│   ├── train.csv                      # Tile IDs for training
│   ├── validation.csv                 # Tile IDs for validation
│   └── test.csv                       # Tile IDs for testing
│
└── scripts/
    ├── process_las.py                 # PDAL pipeline for LAS→DEM
    ├── crop_tiles.py                  # 256x256 patch extraction
    ├── fix_nodata.py                  # NoData standardization
    ├── write_qgis_metadata.py         # QGIS .qmd generator
    └── cleanup_aux_xml.py             # Remove QGIS temp files

📥 Download

Quick Download

# Using zenodo_get (pip install zenodo_get)
zenodo_get 10.5281/zenodo.21229785

🎓 Citation

If you use this dataset in your research, please cite:

@dataset{correa_2026_matchgeo_tdml,
  author       = {Correa, Sabrina Paes Leme P. and Panzini, D. and Oliveira, H. N. and Belton, D. and Iv'{a}nov'{a}, I. and Santos, A. de Paula},
  title        = {{MatchGeo: Digital Elevation Model Dataset for Local Feature Matching (TrainingDML-AI Compliant)}},
  year         = {2026},
  publisher    = {Zenodo},
  version      = {1.3},
  doi          = {10.5281/zenodo.21229785},
  url          = {https://doi.org/10.5281/zenodo.21229785},
  note         = {OGC TrainingDML-AI (23-008r3 / 24-006r1) compliant metadata. Contains data derived from REMA, GeoSampa, OpenTopography, CNIG, Maanmittauslaitos, Geobasis NRW, LINZ, LiPAD, and USGS 3DEP}
}

Plain text citation:
Correa, S. P. L. P., Pazini Pedro, D. F., Oliveira, H. N., Belton, D., Ivánová, I., & Santos, A. de Paula. (2026). MatchGeo: Digital Elevation Model Dataset for Local Feature Matching (Version 1.3) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.21229785

Source Dataset Citations

When using specific cities, also cite the original sources (see DATASET_DESCRIPTION.md Section 8 for full BibTeX).

📜 License & Attribution

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

You are free to:

• Share: Copy and redistribute the material in any medium or format
• Adapt: Remix, transform, and build upon the material for any purpose, even commercially

Under the following terms:

• Attribution: You must give appropriate credit, provide a link to the license, and indicate if changes were made

Required Attribution Statements

When using this dataset, your publication or product must include:

1. Dataset citation (see Citation section)
2. Original source acknowledgments (per region):
   • ATA_MV: Data derived from REMA © Polar Geospatial Center / University of Minnesota
   • BRA_SP: Data derived from GeoSampa © Prefeitura de São Paulo
   • CHN_WS: Data derived from OpenTopography dataset by Zhou, C. (DOI: 10.5069/G98C9TGT)
   • ESP_EH: Data derived from PNOA-LiDAR © CNIG / Instituto Geográfico Nacional
   • FIN_LM: Data derived from Maanmittauslaitos © National Land Survey of Finland
   • GER_BN: Data derived from Geobasis NRW © Bezirksregierung Köln
   • IDN_SV: Data derived from OpenTopography dataset by Carr, B. (DOI: 10.5069/G8988568)
   • KAZ_AC: Data derived from OpenTopography dataset by Amey et al. (DOI: 10.5069/G9H41PMP)
   • KSA_WA: Data derived from OpenTopography dataset by Matthieu et al. (DOI: 10.5069/G9V40SDZ)
   • NAM_HF: Data derived from OpenTopography dataset by Salomon et al. (DOI: 10.5069/G9W957BC)
   • NZL_KP: Data derived from LINZ © Land Information New Zealand
   • PHL_TA: Data derived from LiPAD © UP Diliman TCAGP / DREAM Program
   • USA_GC: Data derived from USGS 3DEP © U.S. Geological Survey

🏗️ Processing Pipeline

All cities were processed through a standardized PDAL pipeline:

1. Acquisition — Raw data from source portals (LAZ, DEM, point clouds)
2. Preprocessing — Region-specific filtering (ground classification, outlier removal, noise filtering)
3. Rasterization — PDAL writers.gdal with output_type=max (DSM)
4. Standardization — Float32, NoData=-9999, BigTIFF, Tiled, DEFLATE compression
5. Patch Extraction — Non-overlapping 256x256 pixel grid
6. Annotation — Handcrafted keypoints (Bonn, São Paulo)
7. Metadata — ISO 19115-2 + OGC TDML compliant per-region metadata

See scripts/process_las.py for the full PDAL pipeline.

🐛 Issues & Support

• Bug reports: GitHub Issues
• Questions: GitHub Discussions
• Email: sabrina.correa@ufv.br

📅 Changelog

v1.3 (2026-07-24)

• Updated annotation structure according to OGC TDML
• Modified metadata to be in accordance to OGC TDML

v1.2 (2026-07-22)

• Adjusted metadata according to FAIR principles
• Added finalised annotation for BRA_SP
• Re-cropped images to 256x256 pixels

v1.1 (2026-05-11)

• Expanded to 13 cities across 6 continents
• Reorganized into data/ folder with per-region metadata
• Added QGIS .qmd metadata files
• Standardized NoData values to -9999.0
• Added BigTIFF, tiled, DEFLATE compression to addapt to COG files
• Updated to MatchGeo branding

v1.0 (2026-03-30)

• Initial release
• Bonn: 20,000+ handcrafted annotations
• Sao Paulo: 10,000+ handcrafted annotations

🙏 Acknowledgments

• Data providers: REMA (Polar Geospatial Center / University of Minnesota), Prefeitura de São Paulo (GeoSampa), OpenTopography, CNIG (Spain), Maanmittauslaitos (Finland), Geobasis NRW (Germany), LINZ (New Zealand), UP Diliman TCAGP / DREAM (Philippines), USGS (United States)
• Imagery providers: CNES / Airbus DS (Pleiades), Maxar (WorldView-3), SPOT Image (SPOT 6)
• Funding: This project is currently funded by CNPq (Brazil)
• Institutional support: Universidade Federal de Viçosa (UFV)

Maintainer: Sabrina Correa | Universidade Federal de Viçosa | sabrina.correa@ufv.br

Last Updated: 2026-07-24