Update README.md

updated readme v00

README.md

@@ -1,3 +1,533 @@
----
-license: cc-by-4.0
----
+# GeoMatch-DEM: Multi-City Digital Elevation Model Dataset for Local Feature Matching
+
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.XXXXXXX.svg)](https://doi.org/10.5281/zenodo.XXXXXXX)
+[![License: CC BY 4.0](https://img.shields.io/badge/License-CC%20BY%204.0-lightgrey.svg)](https://creativecommons.org/licenses/by/4.0/)
+[![Python 3.8+](https://img.shields.io/badge/python-3.8+-blue.svg)](https://www.python.org/downloads/)
+
+**A curated multi-city Digital Elevation Model (DEM) dataset for training and benchmarking local feature matching algorithms in urban terrain analysis.**
+
+---
+
+## 🎯 Overview
+
+GeoMatch-DEM aggregates high-resolution elevation data from three distinct urban environments—Bonn (Germany), São Paulo (Brazil), and Fortaleza (Brazil)—to support research in cross-domain local feature detection and matching. The dataset provides standardized 333×333 pixel patches with handcrafted and automated ground truth annotations for training computer vision models on geospatial data.
+
+### Key Features
+
+- **Multi-source fusion**: LiDAR (Bonn, São Paulo) and photogrammetric (Fortaleza) elevation data
+- **Standardized format**: All cities processed to 333×333 pixel GeoTIFF tiles
+- **Rich annotations**: 27,000+ verified patch pairs from Bonn alone
+- **Cross-city evaluation**: Explicit intra-city and inter-city test splits
+- **FAIR compliant**: ISO 19115-2 metadata, DOI registration, open access
+
+---
+
+## 📊 Dataset Statistics
+
+| City | Source | Method | Resolution | Tiles | CRS | Year | Area |
+|------|--------|--------|------------|-------|-----|------|------|
+| **Bonn** | IEEE GeoNRW | Airborne LiDAR | 1.0 m | [X] | EPSG:25832 | 2016-2018 | City limits |
+| **São Paulo** | GeoSampa | Airborne LiDAR | **0.5 m** | 700+ | EPSG:31983 | [Year] | Municipal |
+| **Fortaleza** | Municipal Survey | Aerial photogrammetry | 1.0 m | ~3,000 | EPSG:31984 | 2016-2017 | 333 km² |
+
+**Ground Truth**: 27,000+ handcrafted annotations (Bonn), plus automated pairs/triplets  
+**Splits**: 80% train / 10% validation / 10% test (stratified by city and difficulty)  
+**Total Size**: ~[X] GB  
+**Patch Size**: 333 × 333 pixels (physical extent: 166.5m × 166.5m for 0.5m data)
+
+---
+
+## 📥 Download
+
+| Repository | Link | Notes |
+|------------|------|-------|
+| **Zenodo (Primary)** | [https://doi.org/10.5281/zenodo.XXXXXXX](https://doi.org/10.5281/zenodo.XXXXXXX) | DOI-backed, permanent archive |
+| Hugging Face Datasets | [https://huggingface.co/datasets/paeslemesa/geomatch-dem](https://huggingface.co/datasets/paeslemesa/geomatch-dem) | Streaming loader available |
+| Academic Torrents | [Link TBD] | For large-scale institutional downloads |
+
+### Quick Download
+
+```bash
+# Using zenodo_get (pip install zenodo_get)
+zenodo_get 10.5281/zenodo.XXXXXXX
+
+# Or wget/curl (specific file)
+wget https://zenodo.org/record/XXXXXXX/files/GeoMatch-DEM-v1.0.zip
+```
+
+---
+
+## 🎓 Citation
+
+If you use this dataset in your research, please cite:
+
+```bibtex
+@dataset{correa_2026_geomatch_dem,
+  author       = {Correa, S. P. L. P. and Santos, A. de Paula and Oliveira, H. N.},
+  title        = {{GeoMatch-DEM: Multi-City Digital Elevation Model Dataset for Local Feature Matching}},
+  year         = 2026,
+  publisher    = {Zenodo},
+  version      = {1.0},
+  doi          = {10.5281/zenodo.XXXXXXX},
+  url          = {https://doi.org/10.5281/zenodo.XXXXXXX},
+  note         = {Contains data derived from IEEE GeoNRW, GeoSampa, and Fortaleza City Hall}
+}
+```
+
+### Original Data Source Citations
+
+You must also cite the original data providers:
+
+**Bonn — IEEE GeoNRW:**
+```bibtex
+@misc{ieee_grss_2020_geonrw,
+  author       = {{IEEE GRSS DASE Committee}},
+  title        = {{IEEE GeoNRW Dataset}},
+  year         = 2020,
+  publisher    = {IEEE DataPort},
+  doi          = {10.21227/nd5j-5q42},
+  url          = {https://dx.doi.org/10.21227/nd5j-5q42}
+}
+```
+
+**São Paulo — GeoSampa:**
+```bibtex
+@misc{prefeitura_sp_geosampa,
+  author       = {{Prefeitura de São Paulo, Secretaria Municipal de Urbanismo e Licenciamento}},
+  title        = {{GeoSampa - Sistema de Informação Geográfica - Dados de Altimetria}},
+  url          = {https://geosampa.prefeitura.sp.gov.br/},
+  note         = {LiDAR data processed for this dataset using PDAL}
+}
+```
+
+**Fortaleza — Municipal Survey:**
+```bibtex
+@misc{fortaleza_2016_mdt,
+  author       = {{Prefeitura Municipal de Fortaleza, SEUMA}},
+  title        = {{Modelo Digital de Terreno - Levantamento Aerofotogramétrico 2016-2017}},
+  year         = 2016,
+  organization = {Esteio Engenharia e Aerolevantamento S/A},
+  url          = {[URL if available]},
+  note         = {Survey dates: 2016-12-20, 23, 27; 2017-01-06}
+}
+```
+
+---
+
+## 🗂️ Dataset Structure
+
+```
+GeoMatch-DEM-v1.0/
+├── README.md                          # This file
+├── CITATION.cff                       # Citation File Format
+├── LICENSE                            # CC BY 4.0 full text
+├── data/
+│   ├── bonn/
+│   │   ├── processed/                 # Standardized GeoTIFF tiles (1m res)
+│   │   │   ├── bonn_001.tif
+│   │   │   └── ...
+│   │   └── annotations/               # Bonn-specific ground truth
+│   │       ├── handcrafted_pairs.csv  # 27,000+ verified annotations
+│   │       └── synthetic_pairs.csv    # Automated augmentations
+│   ├── sao_paulo/
+│   │   ├── processed/                 # 700+ GeoTIFF tiles (0.5m res)
+│   │   │   ├── sp_001.tif
+│   │   │   └── ...
+│   │   └── annotations/
+│   └── fortaleza/
+│       ├── processed/                 # ~3,000 GeoTIFF tiles (1m res)
+│       │   ├── for_001.tif
+│       │   └── ...
+│       └── annotations/
+├── splits/
+│   ├── train.csv                      # 80% of all pairs
+│   ├── val.csv                        # 10% validation
+│   └── test.csv                       # 10% test (hold-out)
+├── metadata/
+│   ├── iso19139.xml                   # ISO 19115-2 compliant metadata
+│   ├── datacite.xml                   # DataCite registration metadata
+│   ├── schemaorg.jsonld               # Schema.org/JSON-LD for web discovery
+│   └── prov/
+│       ├── provenance.ttl             # W3C PROV-O RDF triples
+│       └── processing.log             # Detailed processing history
+└── docs/
+    ├── DATASET_DESCRIPTION.md         # Full technical specifications
+    ├── PREPROCESSING.md               # Reproducible processing pipelines
+    ├── QUALITY_REPORT.md              # Per-city quality assessment
+    └── CITY_COMPARISON.md             # Cross-city characteristics analysis
+```
+
+---
+
+## 🚀 Quick Start
+
+### Installation
+
+```bash
+pip install rasterio pandas numpy matplotlib
+```
+
+### Loading DEM Tiles
+
+```python
+import rasterio
+import numpy as np
+import matplotlib.pyplot as plt
+
+def load_dem(tile_path):
+    """
+    Load a DEM tile and return elevation data with geospatial metadata.
+
+    Parameters
+    ----------
+    tile_path : str or Path
+        Path to GeoTIFF file
+
+    Returns
+    -------
+    dem : ndarray
+        2D array of elevation values (float32)
+    transform : Affine
+        Rasterio affine transformation matrix
+    crs : CRS
+        Coordinate reference system
+    profile : dict
+        Full rasterio profile metadata
+    """
+    with rasterio.open(tile_path) as src:
+        dem = src.read(1)
+        transform = src.transform
+        crs = src.crs
+        profile = src.profile
+
+    # Replace nodata with NaN for visualization
+    if profile.get('nodata') is not None:
+        dem = np.where(dem == profile['nodata'], np.nan, dem)
+
+    return dem, transform, crs, profile
+
+# Example: Load a São Paulo tile
+dem, transform, crs, profile = load_dem('data/sao_paulo/processed/sp_001.tif')
+
+print(f"Tile shape: {dem.shape}")           # (333, 333)
+print(f"CRS: {crs}")                        # EPSG:31983
+print(f"Resolution: {transform.a}m")       # 0.5m
+print(f"Elevation range: [{np.nanmin(dem):.1f}, {np.nanmax(dem):.1f}]m")
+
+# Visualize
+plt.figure(figsize=(8, 8))
+plt.imshow(dem, cmap='terrain', vmin=np.nanpercentile(dem, 2), vmax=np.nanpercentile(dem, 98))
+plt.colorbar(label='Elevation (m)')
+plt.title('São Paulo DEM Sample')
+plt.axis('off')
+plt.show()
+```
+
+### Loading Ground Truth Pairs
+
+```python
+import pandas as pd
+from pathlib import Path
+
+# Load train split
+train_df = pd.read_csv('splits/train.csv')
+
+print(f"Total training pairs: {len(train_df)}")
+print(f"Cities represented: {train_df[['city_a', 'city_b']].stack().unique()}")
+
+# Get a verified matching pair (handcrafted annotation)
+match_row = train_df[(train_df['match_valid'] == 1) & 
+                     (train_df['annotation_type'] == 'handcrafted')].iloc[0]
+
+dem_a, _, _, _ = load_dem(match_row['patch_a_path'])
+dem_b, _, _, _ = load_dem(match_row['patch_b_path'])
+
+print(f"Pair type: {match_row['city_a']} → {match_row['city_b']}")
+print(f"Match valid: {match_row['match_valid']}")
+print(f"Annotation: {match_row['annotation_type']}")
+
+# Display side-by-side
+fig, axes = plt.subplots(1, 2, figsize=(12, 5))
+axes[0].imshow(dem_a, cmap='terrain')
+axes[0].set_title(f"Reference: {Path(match_row['patch_a_path']).stem}")
+axes[1].imshow(dem_b, cmap='terrain')
+axes[1].set_title(f"Match: {Path(match_row['patch_b_path']).stem}")
+for ax in axes:
+    ax.axis('off')
+plt.tight_layout()
+plt.show()
+```
+
+### Loading Triplet Format (for Metric Learning)
+
+```python
+# For models using anchor-positive-negative triplets
+triplet_df = pd.read_csv('splits/train_triplets.csv')
+
+row = triplet_df.iloc[0]
+anchor = load_dem(row['anchor_path'])[0]
+positive = load_dem(row['positive_path'])[0]
+negative = load_dem(row['negative_path'])[0]
+
+# anchor and positive are matching (same location/similar)
+# anchor and negative are non-matching (different location)
+```
+
+---
+
+## 🌍 Data Sources & Processing
+
+### Bonn, Germany — IEEE GeoNRW
+
+**Source**: IEEE GRSS Data and Algorithm Standard Evaluation (DASE) Committee  
+**Method**: Airborne LiDAR survey  
+**Resolution**: 1.0 m  
+**CRS**: ETRS89 / UTM zone 32N (EPSG:25832)  
+**Vertical Datum**: DHHN92 (German Height Reference System)  
+**Year**: 2016–2018  
+
+**Processing Applied**:
+- Downloaded original GeoNRW tiles covering Bonn municipal area
+- Subset to city boundaries using administrative boundary mask
+- Re-tiled to standardized 333×333 pixel patches
+- Format: GeoTIFF, 32-bit float, no compression (for fast I/O)
+- Quality: ±0.1 m vertical accuracy (highest quality tier)
+
+**Annotations**: 27,000+ handcrafted patch pairs for training/validation  
+**Role in Dataset**: High-quality baseline for algorithm development
+
+---
+
+### São Paulo, Brazil — GeoSampa
+
+**Source**: Prefeitura de São Paulo, SMUL (Secretaria Municipal de Urbanismo e Licenciamento)  
+**Method**: Airborne LiDAR (700+ original LAS tiles)  
+**Resolution**: **0.5 m** (highest in dataset)  
+**CRS**: SIRGAS 2000 / UTM zone 23S (EPSG:31983)  
+**Vertical Datum**: SIRGAS 2000  
+**Year**: [Check GeoSampa metadata for exact survey year]  
+
+**Processing Pipeline** (PDAL via Python):
+
+```python
+# Reproducible PDAL pipeline used for processing
+import pdal
+
+pipeline_json = """
+[
+    {
+        "type": "readers.las",
+        "filename": "input.laz"
+    },
+    {
+        "type": "filters.smrf",
+        "scalar": 1.25,
+        "slope": 0.15,
+        "threshold": 0.5,
+        "window": 16.0
+    },
+    {
+        "type": "writers.gdal",
+        "filename": "output.tif",
+        "resolution": 0.5,
+        "output_type": "max",
+        "data_type": "float32",
+        "nodata": -9999
+    }
+]
+"""
+
+pipeline = pdal.Pipeline(pipeline_json)
+pipeline.execute()
+```
+
+**Rationale**:
+- **SMRF (Simple Morphological Filter)**: Selected for ground classification in dense urban environment with mixed topography
+- **Parameters**: Tuned for São Paulo's vertical morphology (flat floodplains + steep hills)
+- **output_type="max"**: Conservative ground estimate using highest ground returns to avoid building residuals
+- **No compression**: Prioritized read speed for deep learning training (dataset stored on fast SSD/NVMe recommended)
+
+**Quality Notes**:
+- Expected vertical accuracy: ±0.15 m (based on GeoSampa LiDAR specifications)
+- Known artifacts: Water bodies (Tietê/Pinheiros rivers), dense vegetation gaps (>95% canopy cover)
+- Coverage gaps: Very dense urban canyons may have incomplete ground returns
+
+---
+
+### Fortaleza, Brazil — Municipal Survey 2016-2017
+
+**Source**: Prefeitura Municipal de Fortaleza, SEUMA (Secretaria de Urbanismo e Meio Ambiente)  
+**Contractor**: Esteio Engenharia e Aerolevantamento S/A  
+**Method**: Aerial photogrammetric survey  
+**Resolution**: 1.0 m  
+**CRS**: SIRGAS 2000 / UTM zone 24S (EPSG:31984)  
+**Vertical Datum**: [Local datum - verify in source]  
+**Survey Dates**: 20, 23, 27 December 2016; 06 January 2017  
+
+**Coverage**:
+- **Total Area**: 333 km² (full municipal coverage)
+- **Detailed Area**: 20 km² high-density core at enhanced resolution
+- **Detailed Neighborhoods**: Mucuripe, Meireles, Varjota, Aldeota, Dionísio Torres, Joaquim Távora, plus partial: Papicu, Vicente Pinzon, Cocó, São João do Tauape, Fátima, José Bonifácio, Centro, Praia de Iracema, Cais do Porto
+
+**Processing**:
+- Pre-processed DEM delivered by contractor
+- Ground filtering applied by Esteio Engenharia (methodology: proprietary)
+- Point density: 1 point/m² (general area), higher in detailed zone
+- Subset to 333×333 pixel tiles matching dataset standard
+
+**Quality Notes**:
+- Expected vertical accuracy: ±0.3 m (typical for 1:1000 photogrammetry)
+- **Important**: 2016-2017 acquisition date—significant urban development since then
+- Limitations: Photogrammetric noise higher than LiDAR; vegetation filtering less reliable
+- Coastal areas: Possible tide-related artifacts in beach zones
+
+---
+
+## 🏷️ Ground Truth & Annotations
+
+### Annotation Types
+
+| Type | Count | Method | Purpose |
+|------|-------|--------|---------|
+| **Handcrafted** | 27,000+ (Bonn) | Manual expert verification | High-confidence training seed, benchmark evaluation |
+| **Synthetic** | [X] | Geometric transforms (rotation, scale, translation) | Data augmentation, invariance training |
+| **Automated** | [X] | SIFT+RANSAC candidate generation | Scale training data |
+
+### CSV Schema
+
+**Pair Format** (`splits/*.csv`):
+
+| Column | Type | Description |
+|--------|------|-------------|
+| `patch_a_path` | str | Relative path to first patch |
+| `patch_b_path` | str | Relative path to second patch |
+| `match_valid` | int (0/1) | 1 = geometrically matching, 0 = negative pair |
+| `city_a` | str | Source city of patch A (`bonn`, `sao_paulo`, `fortaleza`) |
+| `city_b` | str | Source city of patch B |
+| `overlap_percent` | float | Geographic overlap for intra-city pairs |
+| `transform_matrix` | str | JSON 3×3 homography (for synthetic pairs) |
+| `annotation_type` | str | `handcrafted`, `synthetic`, or `automated` |
+| `difficulty` | str | `easy`, `medium`, `hard` (for handcrafted) |
+| `split` | str | `train`, `val`, or `test` |
+
+**Triplet Format** (if applicable):
+- `anchor_path`, `positive_path`, `negative_path`
+- Same metadata columns as above
+
+### Cross-City Strategy
+
+- **Intra-city pairs** (70%): Both patches from same city (baseline difficulty)
+- **Inter-city pairs** (30%): Patches from different cities (tests generalization)
+- **Hard negatives**: Geographically close but non-matching (e.g., adjacent tiles)
+
+---
+
+## 📋 Data Splits
+
+Stratified random sampling ensuring:
+- Balanced city representation
+- Balanced annotation types
+- No geographic leakage (patches from same source tile kept in same split)
+
+```python
+# Load splits
+train = pd.read_csv('splits/train.csv')      # 80%
+val = pd.read_csv('splits/val.csv')          # 10%
+test = pd.read_csv('splits/test.csv')        # 10%
+
+# Verify stratification
+print(train['city_a'].value_counts())
+print(train['annotation_type'].value_counts())
+```
+
+---
+
+## ⚠️ Quality & Usage Notes
+
+### Per-City Characteristics
+
+| Aspect | Bonn | São Paulo | Fortaleza |
+|--------|------|-----------|-----------|
+| **Acquisition** | LiDAR | LiDAR | Photogrammetry |
+| **Accuracy** | ±0.1 m | ±0.15 m | ±0.3 m |
+| **Terrain** | Temperate, rolling | Tropical, flat+steep | Coastal, flat |
+| **Urban density** | Medium | Very high | High (coastal) |
+| **Age** | 2016-2018 | [Year] | 2016-2017 (dated) |
+| **Best for** | Baseline quality | High-res features | Domain robustness |
+
+### Known Limitations
+
+1. **CRS Differences**: Each city uses different UTM zones. Users must handle reprojection for cross-city visualization (model training uses normalized patches).
+2. **Resolution Mismatch**: 0.5m (SP) vs 1.0m (Bonn/Fortaleza). Consider resampling or multi-scale training.
+3. **Temporal Gap**: Fortaleza data is 8-9 years old; significant urban change expected.
+4. **Compression**: São Paulo tiles are uncompressed for training speed—expect larger file sizes.
+5. **Water Artifacts**: All cities have elevation artifacts over water bodies (rivers, reservoirs).
+
+### Recommended Usage
+
+- **Training**: Use all cities for domain generalization
+- **Validation**: Use handcrafted Bonn annotations for reliable metrics
+- **Testing**: Hold-out test set includes inter-city pairs for generalization assessment
+- **Domain Adaptation**: Bonn→São Paulo→Fortaleza provides increasing difficulty curve
+
+---
+
+## 📜 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](#-citation) section)
+2. **Original source acknowledgments**:
+   - "Bonn data derived from IEEE GeoNRW Dataset © IEEE GRSS"
+   - "São Paulo data derived from GeoSampa © Prefeitura de São Paulo"
+   - "Fortaleza data derived from Municipal Survey © Prefeitura de Fortaleza"
+
+---
+
+## 🐛 Issues & Support
+
+- **Bug reports**: [GitHub Issues](https://github.com/[your-repo]/GeoMatch-DEM/issues)
+- **Questions**: [GitHub Discussions](https://github.com/[your-repo]/GeoMatch-DEM/discussions)
+- **Email**: [your-email@institution.edu]
+
+### Known Issues
+
+- [ ] Fortaleza download in progress (ETA: [date])
+- [ ] Total dataset size TBD after Fortaleza integration
+- [ ] Zenodo DOI pending registration
+
+---
+
+## 🗺️ Related Resources
+
+- [IEEE GeoNRW Dataset](https://ieee-dataport.org/open-access/geonrw)
+- [GeoSampa Portal](https://geosampa.prefeitura.sp.gov.br/)
+- [FAIR Principles](https://www.go-fair.org/fair-principles/)
+- [ISO 19115-2 Standard](https://www.iso.org/standard/67039.html)
+
+---
+
+## 📅 Changelog
+
+### v1.0 (2026-03-30)
+- Initial release
+- Bonn: 27,000+ handcrafted annotations
+- São Paulo: 700+ tiles processed (0.5m)
+- Fortaleza: Integration in progress
+
+---
+
+**Maintainer**: Sabrina Correa | Universidade Federal de Vicosa | sabrina.correa@ufv.br
+
+**Last Updated**: 2026-03-30