All Tools
T
DataFreeOpen Source
TORCHGEO
Geospatial deep learning with PyTorch
MIT
ABOUT
Geospatial machine learning has a high barrier to entry — satellite and aerial imagery come in non-standard formats (GeoTIFFs with coordinate reference systems, multi-band rasters, temporal sequences), standard CV datasets and transforms don't handle them, and pre-trained models rarely transfer. TorchGeo provides PyTorch Dataset and DataLoader classes that natively handle geospatial data: multi-band rasters, geographic sampling (random point sampling, bounding boxes), coordinate transforms, and sensor-specific augmentations. It also bundles pre-trained models on remote sensing benchmarks.
INSTALL
pip install torchgeoINTEGRATION GUIDE
1. Train a land cover classification model on satellite imagery with proper geographic sampling and augmentation
2. Detect changes in satellite images over time using temporal sequences of multi-spectral data
3. Fine-tune pre-trained remote sensing models (ResNet, Swin) on custom aerial photography datasets
4. Build a crop type mapping pipeline that integrates with Sentinel-2 or Landsat data feeds
5. Create geospatial embeddings for similarity search across large satellite image archives
TAGS
pythonpytorchgeospatialsatellite-imageryremote-sensinggis