This is a collection of example models implementing the Machine Learning Model Extension to the SpatioTemporal Asset Catalog (STAC) spec. Each metadata json describes a corresponding model asset and the requirements to run that model.
These examples, and
APIs built on top of them, show the utility of describing data linead and runtime requirements of ML models.
The eventual goal is that most geospatial ML models are represented by MLM metadata, making it easy to run them on STAC datasets and derive more value.
See the MLM spec description if you want to learn more about the MLM description fields:
https://github.com/crim-ca/mlm-extensionAnd also check out the stac-model package for creating and validating MLM metadata:
https://github.com/crim-ca/mlm-extension/blob/main/README_STAC_MODEL.mdEach of these models is hosted and deployed in
WherobotsAI Raster Inference, a tool for scaling ML models to planet-scale inference tasks.
Big thanks to
SATLAS and
TorchGeo for distributing open source code and weights for these models.
