TileNet#

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

TileNet (Tile2Vec) encoder.

Added in version 0.9.

Parameters:

  • weights (TileNet_Weights | None) – Pre-trained TileNet weights to load.

  • *args (Any) – Positional arguments.

  • **kwargs (Any) – Keyword arguments forwarded to model.

Returns:

A TileNet model.

Return type:

Module

class torchgeo.models.TileNet(in_channels=4, z_dim=512)[source]#

Bases: Module

TileNet encoder.

versionadded:: 0.9

__init__(in_channels=4, z_dim=512)[source]#

Initialize TileNet.

Parameters:

  • in_channels (int) – Number of input channels.

  • z_dim (int) – Output embedding dimension.

forward(x)[source]#

Compute TileNet embeddings.

Parameters:

x (Tensor) – Input image tensor of shape (B, C, H, W).

Returns:

Embedding tensor of shape (B, embedding_dim).

Return type:

Tensor

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

Bases: WeightsEnum

TileNet (Tile2Vec) weights.

NAIP-pretrained Tile2Vec encoder.

Added in version 0.9.

__new__(value)#