TorchGeo: PyTorch-Native Geospatial Datasets and Remote Sensing Models

TorchGeo provides a collection of geospatial datasets, samplers, and pre-trained models for earth observation and remote sensing tasks. Developed by the TorchGeo organization and updated through March 2026, it standardizes access to satellite imagery within the PyTorch ecosystem. The library facilitates the integration of diverse geospatial data sources into deep learning workflows.

Use Cases

• Land cover classification using multi-spectral bands from integrated datasets like EuroSAT
• Semantic segmentation of infrastructure using high-resolution satellite imagery and land cover labels
• Spatial sampling of large-scale raster data using geographic coordinate-aware samplers

Strengths

• MIT License allows for open commercial and academic application
• Includes pre-trained models specifically optimized for remote sensing imagery
• Provides specialized transforms for multi-spectral data not available in standard vision libraries

Limitations

• Data availability is dependent on the uptime and hosting status of external third-party providers
• High computational and memory requirements for processing large-scale raster imagery

Provenance

Source

TorchGeo GitHub repository and various earth observation agencies

Collection Method

Aggregated and standardized from multiple remote sensing data providers

Freshness

Updated as of March 2026; frequency depends on individual dataset integrations.

Geography

Global