API Reference

Core package

Utility helpers

rshf.utils.from_config(model_class, repo_id, revision=None, **kwargs)[source]

Load a model with randomly initialized weights using the architecture configuration stored in a HuggingFace Hub repository.

This is useful for training a model from scratch while still using the same architecture as a known pretrained checkpoint.

Parameters:

  • model_class – The model class to instantiate (e.g. SatMAE).

  • repo_id (str) – HuggingFace Hub repository ID (e.g. "MVRL/satmae-vitlarge-fmow-pretrain-800").

  • revision (str, optional) – Branch, tag, or commit hash to use. Defaults to the latest revision.

  • **kwargs – Additional keyword arguments that override values read from the repository’s config.json. These must be valid parameters for model_class.__init__; unknown parameters will raise an error when the model is instantiated.

Returns:

An instance of model_class with randomly initialized weights.

Raises:

  • huggingface_hub.utils.EntryNotFoundError – If config.json is not found in the repository.

  • huggingface_hub.utils.RepositoryNotFoundError – If repo_id does not exist or is not accessible.

Example

>>> from rshf import from_config
>>> from rshf.satmae import SatMAE
>>> model = from_config(SatMAE, "MVRL/satmae-vitlarge-fmow-pretrain-800")
rshf.utils.help(model)[source]
rshf.utils.list_models(model_name)[source]

Model entry-point modules

The following modules expose the primary model classes:

  • rshf.bioclip

  • rshf.climplicit

  • rshf.clip

  • rshf.croma

  • rshf.dinov3_sat

  • rshf.geoclap

  • rshf.geoclip

  • rshf.presto

  • rshf.prithvi

  • rshf.prom3e

  • rshf.rcme

  • rshf.remoteclip

  • rshf.rvsa

  • rshf.sat2cap

  • rshf.satclip

  • rshf.satmae

  • rshf.satmaepp

  • rshf.scalemae

  • rshf.senclip

  • rshf.sinr

  • rshf.streetclip

  • rshf.taxabind

See Model loading guide for load snippets and Functions and helpers for helper functions.