Utilities

The signxai.utils module provides utility functions used by both the PyTorch and TensorFlow implementations.

General Utilities

signxai.utils.load_image(path, target_size=(224, 224), preprocess=True)

Loads and preprocesses an image for use with models.

Parameters:

  • path (str) – Path to the image file

  • target_size (tuple, optional) – Target size for the image (height, width)

  • preprocess (bool, optional) – Whether to preprocess the image (ImageNet normalization)

Returns:

Tuple of (original image, preprocessed image)

Return type:

tuple

signxai.utils.normalize_heatmap(heatmap, percentile=99)

Normalizes a heatmap to the specified percentile for visualization.

Parameters:

  • heatmap (numpy.ndarray) – The heatmap to normalize

  • percentile (float, optional) – The percentile value for normalization

Returns:

Normalized heatmap

Return type:

numpy.ndarray

signxai.utils.download_image(path, url=None)

Downloads an example image if it doesn’t exist.

Parameters:

  • path (str) – Path to save the image

  • url (str, optional) – URL to download from (default: example image)

Returns:

None

signxai.utils.download_model(path, url=None)

Downloads an example model if it doesn’t exist.

Parameters:

  • path (str) – Path to save the model

  • url (str, optional) – URL to download from (default: example model)

Returns:

None

signxai.utils.ensure_dir(file_path)

Ensures that a directory exists.

Parameters:

file_path (str) – Path to check/create

Returns:

None

signxai.utils.remove_softmax(model)

Removes the softmax activation from a model.

Parameters:

model – Model to process

Returns:

Model with softmax removed

Raises:

NotImplementedError: If the model framework is not supported

Visualization Utilities

signxai.utils.plot_relevancemap(relevance_map, ax=None, colorbar_ax=None, colorbar_kw=None, **kwargs)

Plots a relevance map as a heatmap.

Parameters:

  • relevance_map (numpy.ndarray) – The relevance map to visualize

  • ax (matplotlib.axes.Axes, optional) – Matplotlib axes to plot on

  • colorbar_ax (matplotlib.axes.Axes, optional) – Axes for the colorbar

  • colorbar_kw (dict, optional) – Additional keyword arguments for the colorbar

  • kwargs – Additional keyword arguments for imshow

Returns:

Matplotlib image object

Return type:

matplotlib.image.AxesImage

signxai.utils.plot_comparison(original_image, explanations, method_names, figsize=(15, 6), cmap='seismic')

Plots multiple explanation methods side by side for comparison.

Parameters:

  • original_image (numpy.ndarray) – The original input image

  • explanations (list) – List of explanations to compare

  • method_names (list) – Names of the methods for the explanations

  • figsize (tuple, optional) – Figure size

  • cmap (str, optional) – Colormap for the heatmaps

Returns:

Matplotlib figure

Return type:

matplotlib.figure.Figure

Data Handling

signxai.utils.batch_to_numpy(batch)

Converts a batch of tensors to numpy arrays.

Parameters:

batch (torch.Tensor or tf.Tensor or numpy.ndarray) – Batch of tensors

Returns:

Batch as numpy array

Return type:

numpy.ndarray

signxai.utils.ensure_batch_dimension(x)

Ensures input has a batch dimension.

Parameters:

x (numpy.ndarray) – Input tensor

Returns:

Input with batch dimension

Return type:

numpy.ndarray

Framework-Specific Utilities

TensorFlow Utilities

signxai.utils.calculate_explanation_innvestigate(model, x, method, **kwargs)

Interface to iNNvestigate for explanation generation.

Parameters:

  • model (tf.keras.Model) – TensorFlow model

  • x (numpy.ndarray) – Input tensor

  • method (str) – iNNvestigate method name

  • kwargs – Additional parameters for the method

Returns:

Explanation

Return type:

numpy.ndarray

PyTorch Utilities

signxai.utils.numpy_to_torch(array, requires_grad=True)

Converts a numpy array to a PyTorch tensor.

Parameters:

  • array (numpy.ndarray) – Numpy array

  • requires_grad (bool, optional) – Whether the tensor requires gradients

Returns:

PyTorch tensor

Return type:

torch.Tensor

signxai.utils.torch_to_numpy(tensor)

Converts a PyTorch tensor to a numpy array.

Parameters:

tensor (torch.Tensor) – PyTorch tensor

Returns:

Numpy array

Return type:

numpy.ndarray