EMCqMRI.core.base.base_dataset module

class EMCqMRI.core.base.base_dataset.Dataset(config_object)[source]

Bases: abc.ABC

Base class for implementation of datasets :param config_object: Configuration object where following attributes

must be specified:

args.engine.fileExtension ([string])

abstract get_label(idx)[source]

Abstract function for fetching labels and masks from the data directory.

Need subclass to implement different logics, like Relaxometry, Reconstruction, QSM, etc.

Parameters: idx ([int]) – Index automatically provided by the DataLoader. It ranges from 0 to len(self.filename_list)-1.
Raises: NotImplementedError – When the subclass does not override this method.
Returns: The subclass implementation should return the label, and, if available, a mask. For seemless integration with the framework, label should have shape [N, X, Y, …] and the mask should have shape [1,X,Y, …], where N is the number of channels (e.g. weighted images) and X,Y,… are the image dimensions.
Return type: (Label [torch.Tensor]), (Mask [torch.Tensor] - optional)

get_length()[source]

It returns the length of the dataset (i.e. number of files in the dataset folder).

abstract get_signal(idx)[source]

Abstract function for fetching the signal from the data directory.

Need subclass to implement different logics, like Relaxometry, Reconstruction, QSM, etc.

Parameters: idx ([int]) – Index automatically provided by the DataLoader. It ranges from 0 to len(self.filename_list)-1.
Raises: NotImplementedError – When the subclass does not override this method.
Returns: The subclass implementation should return the signal. For seemless integration with the framework, signal should have shape [N, X, Y, …], where N is the number of channels (e.g. weighted images) and X,Y,… are the image dimensions.
Return type: (Signal [torch.Tensor])

index_files()[source]

Index all files within the folder path contained in self.path.

If fileExtension is provided (as an option in engine_configuration.txt), only files that match this file extension will be indexed.

load_file(idx)[source]

It loads the file indexed by idx.

It is called when the option preLoadData==False.

load_folder()[source]

It pre-loads all the data in self.path.

It is called when the option preLoadData==True. Use with care, if the size of the data is too large, might cause memory issues.

set_data_path(path)[source]

Set the path for the folder containing training, validation or testing files.

Parameters

path ([string]) – path for training or testing data. It is provided as an option in engine_configuration.txt.

Raises