Back to Gap Map Infrastructure / Classic ML & computer vision

TorchGeo

TorchGeo
open source / Overall score: 2.4

TorchGeo is a PyTorch domain library for geospatial data, modeled on torchvision. It provides remote-sensing datasets, samplers that draw patches by geographic location, transforms, and model weights pretrained on multispectral satellite imagery such as Sentinel-2, so a model can take more bands than RGB. Lightning data modules and tasks for classification, regression and semantic segmentation make published results reproducible.

Openness

5 high confidence
5.0
license
MIT (OSI)
source
public (github.com/torchgeo/torchgeo)
core features withheld
no — maintainer-governed community project; nothing sold
pretrained-weights-terms
per-weight licenses, among them CC-BY-4.0, Apache-2.0, MIT, ODC-BY and OpenRAIL (optional pretrained weights the user selects; they do not govern the MIT library)

The library is MIT-licensed, builds from the public repository and is run by its maintainers by consensus, with nothing sold. Its pretrained weights come from many research groups under their own licenses, but a user picks them one by one, and none is needed to use the library.

Adoption

2 high confidence
2.0

Measured on monthly PyPI downloads of the torchgeo package.

Capability

3 high confidence
3.0

TorchGeo bundles the datasets, pretrained backbones and training tasks to fit and compare models on satellite imagery, as torchvision does for ordinary photographs. Its pretrained weights are starting points for fine-tuning, so a prediction on a new labeling task still needs a training run.

  • https://docs.torchgeo.org/en/stable/api/models.html recorded 2026-11-12

    Models API page: pretrained weights grouped under "Sensor-Agnostic Landsat NAIP Sentinel-1 Sentinel-2 Aerial Atmospheric"; "These weights can be used with imagery from any satellite/sensor" for the sensor-agnostic group.

  • https://raw.githubusercontent.com/torchgeo/torchgeo/HEAD/README.md recorded 2026-11-12

    README: "TorchGeo is a PyTorch domain library, similar to torchvision, providing datasets, samplers, transforms, and pre-trained models specific to geospatial data"; Lightning datamodules "and tasks for various tasks like classification, regression, and semantic segmentation".

Verified 2026-11-12