API Reference: Spec¶

masterful.enums.Task¶

class masterful.enums.Task(value)¶

Bases: enum.Enum

An enum to semantically specify a model’s use case.

Parameters: value – Overriden from enum.Enum. Returns the member of this enum from the corresponding value.

CLASSIFICATION¶: Normal classification task like Alexnet on Imagenet.

BINARY_CLASSIFICATION¶: Binary classification task.

MULTILABEL_CLASSIFICATION¶: Multi-label classification task.

DETECTION¶: Object detection (localization + classification) task.

LOCALIZATION¶: Object localization task.

SEMANTIC_SEGMENTATION¶: Semantic segmentation task.

INSTANCE_SEGMENTATION¶: Instance segmentation task.

KEYPOINT_DETECTION¶: Keypoint detection task.

masterful.enums.ImageRange¶

class masterful.enums.ImageRange(value)¶

Bases: enum.Enum

An enum to model the image input ranges Masterful supports.

ImageRange describes the range of the image pixel values. Common ranges include [0,255] and [0,1]. Some models also prefer to normalize the input data around the ImageNet mean and standard deviation.

Parameters: value – Overriden from enum.Enum. Returns the member of this enum from the corresponding value.

ZERO_ONE¶: Image range is [0,1].

NEG_ONE_POS_ONE¶: Image range is [-1, 1].

ZERO_255¶: Image range is [0,255].

IMAGENET_CAFFE_BGR¶: Image is in BGR channel format, and each channel has been zero-centered around the Imagenet mean, without scaling.

IMAGENET_TORCH¶: Image pixels were scaled to [0,1], then each channel was zero-centered around the Imagenet mean.

CIFAR10_TORCH¶: Image pixels were scaled to [0,1], then each channel was zero-centered around the CIFAR10 mean.

CIFAR100_TORCH¶: Image pixels were scaled to [0,1], then each channel was zero-centered around the CIFAR100 mean.

masterful.enums.TensorStructure¶

class masterful.enums.TensorStructure(value)¶

Bases: enum.Enum

An enum to specify tensor structures.

A tensor structure is the physical data structure used to encapsulate individual tensors. This can either be a single tensor itself, in which case the single tensor is passed to each output of the model, or a tuple of tensors, in which case the number of tensors in the tuple must match exactly the number of outputs in the model.

SINGLE_TENSOR¶: Label or prediction consists of a single tensor, which is passed to all outputs of the model.

TUPLE¶: Label or prediction consists of a tuple of tensors, which must map exactly to the number of outputs of a model.

DICT¶: Label or prediction is a dictionary output tensors.

masterful.enums.BoundingBoxFormat¶

class masterful.enums.BoundingBoxFormat(value)¶

Bases: enum.Enum

Bounding box formats support by Masterful.

TENSORFLOW¶: The Tensorflow bounding box format is (ymin, xmin, ymax, xmax), normalized by the image dimensions into the range [0,1].

VOC¶: The Pascal VOC bounding box format is (xmin, ymin, xmax, ymax) in pixel coordinates.

COCO¶: The MSCOCO bounding box format is (xmin, ymin, width, height) in pixel coordinates.

ALBUMENTATIONS¶: The Albumentations bounding box format is (xmin, ymin, xmax, ymax), normalized by the image dimension into the range [0,1]

YOLO¶: The Yolo bounding box format is (x_center, y_center, width, height), normalized by the image dimensions into the range [0,1]