multigeom.encoders.load_encoder

multigeom.encoders.load_encoder(name_or_path, area=None, *, device=None, repo_id='steevelaquitaine/multigeom', revision=None)

Load an encoder that is already fitted.

• load_encoder(“inception”, area=”V1”): the encoder of the paper (“inception” or “resnet50”, area “V1” or “V4”), downloaded from the Hugging Face repository repo_id the first time and cached after that. The pretrained backbone weights come from torchvision.

• load_encoder(“path/to/encoder.pt”): a checkpoint on disk, saved by NeuralEncoder.save or by the pipeline.

Parameters:

• name_or_path (str | Path)

• area (str | None)

• device (str | torch.device | None)

• repo_id (str)

• revision (str | None)

Return type:

NeuralEncoder