PyTorch domain library for remote sensing
Geospatial deep learning, without the glue code.
Geospatial deep learning isn't just computer vision with larger images — satellite data is a distinct modality with its own geometry, statistics, and metadata.
TorchGeo provides CRS-aware datasets, spatial samplers, multispectral transforms, and pretrained models for satellite and aerial imagery — all as standard PyTorch primitives.
pip install torchgeoMIT · Python 3.12+ · OSGeo projectfrom torchgeo.datasets import EuroSAT
from torchgeo.samplers import RandomGeoSampler
from torchgeo.models import ResNet18_Weights, resnet18
dataset = EuroSAT(root="./data", download=True)
sampler = RandomGeoSampler(dataset, size=256, length=1000)
weights = ResNet18_Weights.SENTINEL2_ALL_MOCO
model = resnet18(weights=weights)Library surface
torchgeo.datasetsGeo-referenced raster and vector datasets with CRS, metadata, and download handling built in.torchgeo.samplersSpatially-aware samplers that iterate over geographic extents, not just array indices.torchgeo.transformsKornia-compatible augmentations for arbitrary band counts, not RGB-only.torchgeo.modelsSatellite-pretrained backbones and task modules that drop into standard PyTorch loops.
Example datasets
