Presto#

class torchgeo.models.Presto(band_groups=None, encoder_embedding_size=128, channel_embed_ratio=0.25, month_embed_ratio=0.25, encoder_depth=2, mlp_ratio=4, encoder_num_heads=8, decoder_embedding_size=128, decoder_depth=2, decoder_num_heads=8, max_sequence_length=24)[source]#

Bases: Module

Pretrained Remote Sensing Transformer (Presto).

Added in version 0.9.

__init__(band_groups=None, encoder_embedding_size=128, channel_embed_ratio=0.25, month_embed_ratio=0.25, encoder_depth=2, mlp_ratio=4, encoder_num_heads=8, decoder_embedding_size=128, decoder_depth=2, decoder_num_heads=8, max_sequence_length=24)[source]#

Initialize a new Presto instance.

Parameters:

  • band_groups (dict[str, Sequence[int]] | None) – Mapping of band group names to channel indices.

  • encoder_embedding_size (int) – Size of the embedding for each token in the encoder.

  • channel_embed_ratio (float) – Ratio of the embedding size to use for channel embeddings in the encoder.

  • month_embed_ratio (float) – Ratio of the embedding size to use for month embeddings in the encoder.

  • encoder_depth (int) – Number of Transformer blocks in the encoder.

  • mlp_ratio (int) – Ratio of the hidden dimension in the MLP compared to the embedding size in the encoder.

  • encoder_num_heads (int) – Number of attention heads in each Transformer block in the encoder.

  • decoder_embedding_size (int) – Size of the embedding for each token in the decoder.

  • decoder_depth (int) – Number of Transformer blocks in the decoder.

  • decoder_num_heads (int) – Number of attention heads in each Transformer block in the decoder.

  • max_sequence_length (int) – Maximum length of the input sequence.

forward(x, dynamic_world, latlons, mask=None, month=0)[source]#

Forward pass of the Presto model.

Parameters:

  • x (Tensor) – Input tensor of shape [batch, timesteps, channels].

  • dynamic_world (Tensor) – Dynamic world tensor of shape [batch, timesteps].

  • latlons (Tensor) – Latitude and longitude tensor of shape [batch, 2].

  • mask (Tensor | None) – Mask tensor of shape [batch, timesteps, channels]. Defaults to None.

  • month (Tensor | int) – Month tensor or integer representing the month. Defaults to 0.

Returns:

Tuple containing the reconstructed inputs for each channel group and the dynamic world output.

Return type:

tuple[Tensor, Tensor | None]

torchgeo.models.presto(weights=None, *args, **kwargs)[source]#

Presto model.

If you use this model in your research, please cite the following paper:

Added in version 0.9.

Parameters:

  • weights (Presto_Weights | None) – Pre-trained model weights to use.

  • *args (Any) – Additional arguments to pass to Presto.

  • **kwargs (Any) – Additional keyword arguments to pass to Presto.

Returns:

A Presto model.

Return type:

Presto

class torchgeo.models.Presto_Weights(*values)[source]#

Bases: WeightsEnum

Presto weights.

Added in version 0.9.

__new__(value)#