Utils

MapImage(image: Union[numpy.ndarray, torch.Tensor], original_values: List[int], reverse: bool) → Union[numpy.ndarray, torch.Tensor]

Maps the current values of a given input image to the values given by the tuple (current values, new values).

Parameters

  • image (Union[np.ndarray, torch.Tensor]) – The input image to transform

  • original_values (List[int]) – List of original values to be mapped

  • reverse (bool) – True if mapping the other way around back to original values

Raises

TypeError – If the input image is neither a numpy array or a torch tensor

Returns

the transformed input after mapping.

Return type

Union[np.ndarray, torch.Tensor]

Example::

transformed_image = MapImage(image, [0, 85, 170]) The values will be mapped to [0, 1, 2] with the first value of the original values specified being the background index.

common_files(folder1, folder2, prefix='Annotation')

Compare two folders and return a list of base filenames (ignoring extensions) that exist in both folders. Files in folder2 may have a prefix (default: ‘Annotation’) that is ignored.

contrast_img(img: numpy.ndarray) → numpy.ndarray

Applies the Adaptive Equalization or histogram equalization contrast method. This method enhances the contrast of an image by adjusting the intensity values of pixels based on the distribution of pixel intensities in the image’s histogram.

Parameters

img (np.ndarray) – input image

Returns

contrasted image

Return type

np.ndarray

createBinaryAnnotation(img: Union[numpy.ndarray, torch.Tensor], frg_class: int) → Union[numpy.ndarray, torch.Tensor]

Creates a binary mask out of the prediction result with root as foreground and the rest as background

Parameters

  • img (Union[np.ndarray, torch.Tensor]) – Input image

  • frg_class (int) – value of the foreground class when creating binary segmentation masks

Raises

TypeError – if the input is neither a numpy array or a torch tensor or if the foreground value is wrong.

Returns

binary mask

Return type

Union[np.ndarray, torch.Tensor]

elliptical_crop(img: numpy.ndarray, center_x: int, center_y: int, width: int, height: int) → Tuple[numpy.ndarray, numpy.ndarray]

Crops out an elliptical shape out of the input image and sets the rest as background

Parameters

  • img (np.ndarray) – Input image

  • center_x (int) – Center x coordinate of the ellipse

  • center_y (int) – Center y coordinate of the ellipse

  • width (int) – Width of the wanted ellipse

  • height (int) – Height of the wanted ellipse

Returns

cropped output image

Return type

Tuple[np.ndarray, np.ndarray]

extract_largest_component_bbox_image(img: Union[numpy.ndarray, torch.Tensor], lab: Optional[Union[numpy.ndarray, torch.Tensor]] = None) → Union[numpy.ndarray, Tuple[numpy.ndarray, numpy.ndarray]]

Extract the largest connected component (LCC) of the given image and get the bounding box associated with this LCC. Depending on the input parameters, it can either:

  • Crop the image to the bounding box.

  • Apply a binary mask to the image, maintaining the same shape as the input image, with everything but the LCC set as a black background.

This function can be applied to both the image and the annotation (label) if provided.

Parameters

  • img (Union[numpy.ndarray, torch.Tensor]) – The input image. - Can be a NumPy array or PyTorch tensor with shape (C, H, W) or (B, C, H, W) if a batch dimension is included. - Must not be None; if None, the function raises a ValueError.

  • lab (Union[numpy.ndarray, torch.Tensor], optional) – The label or annotated image. Defaults to None.

Returns

  • If lab is None: Returns the processed image as a NumPy array or PyTorch tensor.

  • If lab is provided: Returns a tuple containing:

    • The processed image (NumPy array).

    • The processed label image (NumPy array).

Return type

Union[numpy.ndarray, Tuple[numpy.ndarray, numpy.ndarray]]

get_biomass(binary_img: numpy.ndarray) → int

Calculate the biomass by counting the number of pixels equal to 1

Parameters

binary_img (np.ndarray) – input image as binary mask

Returns

integer value corresponding to the pixel count or root biomass.

Return type

int

get_image_paths(dir: str) → List[str]

Goes through a folder directory and lists all filepaths

Parameters

dir (str) – folder directory to extract from all the filepaths

Returns

list of all filepaths

Return type

List[str]

get_weights(labels: torch.Tensor, classes: List[int], device: str, include_background=False) → List[float]

Computes the weights of each class in the batch of labeled images

Parameters

  • labels (torch.Tensor) – Batch of labeled imagse with each pixel of an image equal to a class value.

  • classes (List[int]) – List of the classes

  • device (str) – training device should be ‘cuda’ if training on GPU

  • include_background (bool, optional) – Boolean to include or not the background valued as 0 when calculating the weight of the classes in the images. Defaults to False.

Returns

List of the class weights (floats).

Return type

List[float]