A phone book for orbital deep learning

What it does

This repository is a single, exhaustive curated list of datasets for deep learning with satellite and aerial imagery. It links out to hundreds of resources—Sentinel-1 SAR, Sentinel-2 optical, change-detection benchmarks, crop-mapping time series, marine-debris labels, and more—organized by source and task. The maintainer’s stated usage strategy is bluntly practical: open the page and Control+F for the paper or dataset name you need.

The interesting bit

The value is not code but curation. In a field where data is scattered across Radiant MLHub, Zenodo, Kaggle, AWS Open Data, and individual university servers, this acts as a living card catalog. It even catalogs other catalogs, linking to broader hubs like the Microsoft Planetary Computer and Google Earth Engine Data Catalog.

Key highlights

• Hundreds of dataset links covering Sentinel-1/2, aerial, and multi-spectral imagery.
• Task-specific collections: ship detection, wildfire delineation, crop classification, cloud removal, and change detection.
• Links to meta-resources such as awesome-satellite-imagery-datasets and remote-sensing dataset hubs.
• Includes niche entries like marine-debris classes (plastic, driftwood, sea foam) and offshore wind farm locations.
• No installation or dependencies; it is a pure reference document.

Caveats

• Several entries already rely on web.archive.org snapshots, so expect some link archaeology.
• Dataset descriptions are often just a paper title and a one-line tag; you will still need to visit the source to judge fitness.
• The list is long enough that the README explicitly recommends using your browser’s find function to navigate.

Verdict

Anyone training models on orbital or aerial imagery should bookmark this. If you are looking for code, frameworks, or pre-trained weights, look elsewhere—this is strictly a bibliography of data sources.