AnySat
Produces features from Earth-observation imagery of any resolution, scale or sensor combination with one model.
- Field
- Remote sensing
- Method
- Foundation model
- Data it takes
- imagery
- License
- MIT
AnySat is a single encoder for Earth-observation data that differs in resolution, footprint and sensor. It takes aerial imagery down to 0.2 m alongside time series from Sentinel-1, Sentinel-2, Landsat 7 and 8, ALOS-2 and MODIS, and returns features for a whole tile, for each patch, or densely below the patch scale. Task heads such as linear probes are trained on top.
For per-location work, the dense output is the one to use; the fused patch output barely varies within a tile. Its pretraining data is mostly European, which is worth checking against before relying on it elsewhere.
The code and weights are MIT-licensed.
Catalog entry last checked 2026-10-03. All models →
Run AnySat on your data
The lab runs this on its own compute, in a container, with the inputs and parameters recorded alongside the result. Initial scoping conversations are free.
Contact the Lab