datamodules.RGB.utils package

Submodules

datamodules.RGB.utils.functional module

gt_to_int_encoding(matrix: Tensor, class_encodings: Tensor)[source]

Convert ground truth tensor or numpy matrix to one-hot encoded matrix

Parameters:

  • matrix (torch.Tensor) – Image as a tensor of size [C x H x W] (BGR)

  • class_encodings (List[int]) – class encoding so which class (index) has what value (element)

Returns:

integer encoded matrix of size [#C x H x W]

Return type:

torch.Tensor

gt_to_one_hot(matrix: Tensor, class_encodings: Tensor)[source]

Convert ground truth tensor or numpy matrix to one-hot encoded matrix

Parameters:

  • matrix (torch.Tensor or np.ndarray) – float tensor from to_tensor() or numpy array shape (C x H x W) in the range [0.0, 1.0] or shape (H x W x C) BGR

  • class_encodings (List[int]) – List of int Blue channel values that encode the different classes

Returns:

Tensor of size [#C x H x W] sparse one-hot encoded multi-class matrix, where #C is the number of classes

Return type:

torch.LongTensor

datamodules.RGB.utils.image_analytics module

get_analytics(input_path: Path, data_folder_name: str, gt_folder_name: str, train_folder_name: str, get_img_gt_path_list_func: callable, inmem: bool = False, workers: int = 8) → Tuple[Dict[str, Any], Dict[str, Any]][source]

Get the analytics for the dataset. If the analytics file is not complete, it will be computed and saved.

Parameters:

  • workers (int) – The amount of workers to use for the mean/std computation

  • inmem (bool) – If the dataset should be loaded fully into memory

  • input_path (Path) – Path to the dataset folder

  • data_folder_name (str) – Name of the folder that contains the data

  • gt_folder_name (str) – Name of the folder that contains the ground truth

  • train_folder_name (str) – Name of the folder that contains the training data

  • get_img_gt_path_list_func (callable) – Function that returns a list of tuples with the image and gt path

Returns:

get_class_weights(input_folder: Path, workers=4) → ndarray[source]

Get the weights proportional to the inverse of their class frequencies. The vector sums up to 1

Parameters:

  • input_folder (Path) – Path to the dataset folder (see above for details)

  • workers (int) – Number of workers to use for the mean/std computation

Returns:

The weights vector as a 1D array normalized (sum up to 1)

Return type:

ndarray[double]

datamodules.RGB.utils.output_tools module

output_to_class_encodings(output: ndarray, class_encodings: List[Tuple[int]]) → ndarray[source]

This function converts the output prediction matrix to an image like it was provided in the ground truth

Parameters:

  • output (np.ndarray) – output prediction of the network for a full-size image, where #C is the number of classes

  • class_encodings (List[Tuple[int]]) – Contains the range of encoded classes

Returns:

numpy array of size [C x H x W] (BGR) with the classes encoded as in the ground truth

Return type:

np.ndarray

save_output_page_image(image_name: str, output_image: ndarray, output_folder: Path, class_encoding: List[Tuple[int]]) → None[source]

Helper function to save the output during testing in the DIVAHisDB format

Parameters:

  • image_name (str) – name of the image that is saved

  • output_image (np.ndarray) – output image at full size [#C x H x W]

  • output_folder (Path) – path to the output folder for the test data

  • class_encoding (List[Tuple[int]]) – list with the class encodings

datamodules.RGB.utils.single_transform module

class IntegerEncoding(class_encodings)[source]

Bases: object

class OneHotEncoding(class_encodings)[source]

Bases: object

Module contents