models.fusion.stn package

Submodules

models.fusion.stn.model module

Spatial transformer pytorch modules.

class models.fusion.stn.model.SpatialTransformer(in_channels: int, input_is_3d: bool = False)

Bases: Module

Spatial transformer module.

forward(x: Tensor) → Tensor

Forward pass for the spatial transformer.

Parameters:

x – Image tensor of shape: (N, C, H, W)

Returns:

Image tensor of shape (N, C, H, W)

Return type:

Tensor

class models.fusion.stn.model.STN(transformer_type: SpatialTransformerType, num_param: int, N: int, xdim: int | None = None, **kwargs)

Bases: Module

Base class for a Spatial Transformer Network (STN).

Adapted from: https://github.com/FrederikWarburg/pSTN-baselines. Implement this for specific tasks, which requires the method init_localiser to be implemented.

init_localiser(**kwargs)

Initialise task-specific localiser.

init_model_weights(**kwargs)

Initialise model weights.

forward(x: Tensor, x_high_res: Tensor | None = None)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

forward_localiser(x: Tensor, x_high_res: Tensor | None = None)

Localisation step for spatial transformer.

Parameters:

  • x – Input tensor.

  • x_high_res – Optional high res version of x to transform on.

compute_theta(x: Tensor)

Compute theta parameter through the localisation step.

Parameters:

x – Input tensor.

models.fusion.stn.utils module

Utilities for the feature fusion modules.

models.fusion.stn.utils.init_transformer(transformer_type: SpatialTransformerType, N: int, num_param: int, xdim: int | None = None)

Initialise the spatial transformer.

Parameters:

  • transformer_type – Type of spatial transformer.

  • N – Number of parallel tracks.

  • num_param – If we use an affine (s, r, tx, ty) or crop (0.5, 1, tx, ty) transformation.

  • xdim – Indicator of time seeries datasets. 1 if timeseries, otherwise 2.

class models.fusion.stn.utils.SpatialTransformerType(*values)

Bases: Enum

Enum for spatial transformer type.

AFFINE = 1

Affine transformer.

DIFFEOMORPHIC = 2

Diffeomorphic transformer.

class models.fusion.stn.utils.AffineTransformer(*args, **kwargs)

Bases: Module

Affine spatial transformer.

forward(x: Tensor, params: Tensor, small_image_shape: int | Tuple[int, int])

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class models.fusion.stn.utils.DiffeomorphicTransformer(N: int, num_param: int, xdim: int)

Bases: Module

Diffeomorphic spatial transformer.

forward(x: Tensor, params: Tensor)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

Module contents

Spatial Transformer Network implementation for fusion models.