mml.core.data_loading.task_attributes

class DataSplit

Bases: StrEnum

Represents parts of a dataset that are loaded together. May be selected joint by a fold number to determine the exact samples that are available for iteration over a TaskDataset.

FULL_TRAIN = 'FULL_TRAIN'

TEST = 'TEST'

TRAIN = 'TRAIN'

UNLABELLED = 'UNLABELLED'

VAL = 'VAL'

__new__(value)

class Keyword

Bases: StrEnum

Keyword labels of a task. Refers e.g. to the shown entities within the images.

ANATOMICAL_STRUCTURES = 'anatomical structures'

ANIMALS = 'animals'

ARTIFICIAL = 'artificial'

BONE = 'bone'

BRAIN = 'brain'

BREAST = 'breast'

BUILDINGS = 'buildings'

CAPSULE_ENDOSCOPY = 'capsule endoscopy'

CATARACT_SURGERY = 'cataract_surgery'

CHARS_DIGITS = 'chars_or_digits'

CHEST = 'chest'

CLE = 'confocal laser endomicroscopy'

COLPOSCOPY = 'colposcopy'

CT_SCAN = 'ct_scan'

DERMATOSCOPY = 'dermatoscopy'

DRIVING = 'driving'

ENDOSCOPIC_INSTRUMENTS = 'endoscopic instruments'

ENDOSCOPY = 'endoscopy'

EYE = 'eye'

FACES = 'faces'

FUNDUS_PHOTOGRAPHY = 'fundus_photography'

GASTROSCOPY_COLONOSCOPY = 'gastroscopy_colonoscopy'

GYNECOLOGY = 'Gynecology'

HANDWRITINGS = 'handwritings'

IMAGE_ARTEFACTS = 'image_artefacts'

INSTRUMENT_COUNT = 'counting endoscopic instruments'

LAPAROSCOPY = 'laparoscopy'

LARYNGOSCOPY = 'laryngoscopy'

MEDICAL = 'medical'

MRI_SCAN = 'mri_scan'

NATURAL_OBJECTS = 'natural_objects'

NEPHRECTOMY = 'Nephrectomy'

SCENES = 'scenes'

TISSUE_PATHOLOGY = 'tissue_pathology'

ULTRASOUND = 'ultrasound'

X_RAY = 'x_ray'

__new__(value)

class License

Bases: StrEnum

License for distribution of a task (data).

CC_0_1_0 = 'CC0 1.0 Universal (CC0 1.0) Public Domain Dedication'

CC_BY_4_0 = 'Creative Commons Attribution 4.0 International'

CC_BY_NC_4_0 = 'Creative Commons Attribution-NonCommercial 4.0 International'

CC_BY_NC_SA_4_0 = 'Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International'

CUSTOM = 'License defined in TaskCreator description'

DATABASE_CONTENTS_LICENSE_1_0 = 'Open Data Commons DbCL v1.0'

MIT = 'Massachusetts Institute of Technology'

UNKNOWN = 'unknown'

__new__(value)

class Modality

Bases: StrEnum

The modalities represent the possible keys of a loaded sample from a dataset. E.g. {‘image’: ‘some/path/file.png’, ‘class’: 3}. Note that while TaskDescription stores the Modality as enum in its samples, the loaded batch will contain the str representations!

BBOX = 'bbox'

CLASS = 'class'

CLASSES = 'classes'

IMAGE = 'image'

KEYPOINTS = 'keypoints'

MASK = 'mask'

SAMPLE_ID = 'sample_id'

SOFT_CLASSES = 'soft_classes'

TASK = 'task'

THREE_D_IMAGE = 'three_d_image'

VALUE = 'value'

VIDEO_CLIP = 'video_clip'

__new__(value)

class RGBInfo

Bases: object

Small dataclass storing information about image channels (mostly mean and std).

__init__(r: float = 0.0, g: float = 0.0, b: float = 0.0) → None

b: float = 0.0

g: float = 0.0

get_rgb() → List[float]

r: float = 0.0

to_list() → List[float]

class Sizes

Bases: object

Small dataclass storing information about the dimensionality of a set of images.

__init__(min_height: int = 0, max_height: int = 0, min_width: int = 0, max_width: int = 0) → None

max_height: int = 0

max_width: int = 0

min_height: int = 0

min_width: int = 0

to_list() → List[int]

class TaskType

Bases: StrEnum

Defines the type of task. Different task types usually require completely different architectures and/or training procedures. This is aligned with the torchvision.models split at https://pytorch.org/docs/stable/torchvision/models.html

CLASSIFICATION = 'classification'

DEFAULT = 'unknown'

DETECTION = 'detection'

MULTILABEL_CLASSIFICATION = 'multilabel_classification'

NO_TASK = 'no_task'

REGRESSION = 'regression'

SEMANTIC_SEGMENTATION = 'semantic_segmentation'

UNKNOWN = 'unknown'

VIDEO_CLASS = 'video_classification'

__new__(value)

requires() → List[List[Modality]]

Returns the necessary modalitie(s) for this kind of task. First list level is OR and second level is AND. So if returns [[A, B], [C, D]] either [A and B] or [C and D] are required.