Datasets:
NestEO: A Modular, Hierarchical, and Extensible AI4EO Dataset Framework
NestEO provides a globally aligned, hierarchically nested framework for organizing Earth Observation data. It enables structured pairing, deep metadata filtering, and scalable dataset access across resolutions, regions, and EO modalities.
Directory Structure
NestEO/
├── grids/ # UTM-aligned hierarchical grid Parquet files
├── metadata_current/ # Precomputed proportions: landcover, climate, region
├── datasets_EO/ # Clipped EO imagery tiles (Sentinel, MODIS, etc.)
├── datasets_AUX/ # Auxiliary datasets (DEM, landcover, OSM)
├── embeddings/ # Precomputed model embeddings (e.g., DINOv2, SigLIP)
├── index_structure/ # Source-to-tile and tile-ID index maps
└── versions/ # Dataset snapshots for version control
How to Use
Load a grid from grids/ (e.g., grid_1200m.parquet).
Join with metadata (e.g., metadata_lc.parquet) on tile_id.
Apply spatial or semantic filtering (e.g., only urban, low-cloud).
Select and load EO imagery and AUX sources from paired folders.
Use embeddings (optional) for transfer learning or clustering.
Current Status
This is an initial prototype release. Some EO source folders are partially populated, and ongoing ingestion is underway. Metadata filtering and source-pairing are fully supported.
License
CCBy 4.0.
Contributing
We welcome contributions on:
Region-specific or resolution-specific imagery
Auxiliary or annotation layers
Grid-level metadata enrichment
Benchmarks and model evaluations
vbnet
Copy
Edit