torchgeo/torchgeo ? reverse-engineered prompt

Build me a Python library for geospatial deep learning, kind of like torchvision but for satellite and remote sensing data.

I want it to make working with maps, aerial images, and other geospatial datasets feel easy, even when the files have different coordinate systems, resolutions, or band counts. It should include ready to use dataset wrappers, ways to combine datasets, samplers that can pull image patches from huge scenes, and transforms for common training workflows. Also add a few simple pre trained models or task helpers so someone can get started on things like classification or segmentation without building everything from scratch.

Please make it installable, well documented, and tested, with examples showing how to load geospatial data into a normal PyTorch training loop. If you need to check current docs or package behavior online, go ahead and do that.