FeatureSpec

See the guides: Dataset

class tflibs.datasets.feature_spec.FeatureSpec(shape)

A class for specifying feature specs of a tflibs.datasets.BaseDataset.

feature_proto(value_dict)

Returns a dict of tf.train.Feature proto corresponding to feature_proto_spec

Parameters:value_dict (dict) – A dict containing values of feature_proto_spec Returns:A dict of tf.train.Feature Return type:dict

feature_proto_spec

A property for specifying inner encoding spec of the feature

Returns:The dict containing shape and dtype info Return type:dict

parse(parent_key, record)

Parse TF-record and returns dict of tf.Tensor

Parameters:

• parent_key (str) – The key of the feature
• record (tf.Tensor) – String tensor of TF-record

Returns:

A dict of tensors as specified at feature_proto_spec

Return type:

dict

shape

class tflibs.datasets.feature_spec.IDSpec

Bases: tflibs.datasets.feature_spec.FeatureSpec

A class for specifying unique ID.

create_with_string(string)

feature_proto_spec

A property for specifying inner encoding spec of the feature

Returns:The dict containing shape and dtype info Return type:dict

parse(parent_key, record)

Parse TF-record and returns tf.Tensor

Parameters:

• parent_key (str) – The key of the feature
• record (tf.Tensor) – String tensor of TF-record

Returns:

A scalar tensor containing an id

Return type:

tf.Tensor

class tflibs.datasets.feature_spec.ImageSpec(image_size)

Bases: tflibs.datasets.feature_spec.FeatureSpec

A class for specifying image spec

Parameters:image_size (list|tuple) – The sizes of images

create_with_contents(contents)

create_with_path(path)

create_with_tensor(tensor)

feature_proto_spec

A property for specifying inner encoding spec of the feature

Returns:The dict containing shape and dtype info Return type:dict

parse(parent_key, record)

Parse TF-record and returns tf.Tensor

Parameters:

• parent_key (str) – The key of the feature
• record (tf.Tensor) – String tensor of TF-record

Returns:

A 3-D tensor containing an image

Return type:

tf.Tensor

class tflibs.datasets.feature_spec.LabelSpec(depth, class_names=None)

Bases: tflibs.datasets.feature_spec.FeatureSpec

A class for specifying one-hot label spec

Parameters:

• depth (int) – The number of labels
• class_names (list) – A list of str which describes each labels

create_with_index(index)

create_with_label(label)

feature_proto_spec

A property for specifying inner encoding spec of the feature

Returns:The dict containing shape and dtype info Return type:dict

classmethod from_class_names(class_names)

Create LabelSpec object from a list of names specifying each classes

Parameters:class_names (list) – A list of names specifying each classes Returns:LabelSpec object Return type:LabelSpec

parse(parent_key, record)

Parse TF-record and returns tf.Tensor

Parameters:

• parent_key (str) – The key of the feature
• record (tf.Tensor) – String tensor of TF-record

Returns:

An 1-D tensor containing label

Return type:

tf.Tensor

class tflibs.datasets.feature_spec.MultiLabelSpec(depth, class_names=None)

Bases: tflibs.datasets.feature_spec.FeatureSpec

A class for specifying multi-label spec

Parameters:

• depth (int) – The number of labels
• class_names (list) – A list of str which describes each labels

create_with_labels(labels)

create_with_tensor(tensor)

feature_proto_spec

A property for specifying inner encoding spec of the feature

Returns:The dict containing shape and dtype info Return type:dict

classmethod from_class_names(class_names)

Create MultiLabelSpec object from a list of names specifying each classes

Parameters:class_names (list) – A list of names specifying each classes Returns:MultiLabelSpec object Return type:MultiLabelSpec

parse(parent_key, record)

Parse TF-record and returns tf.Tensor

Parameters:

• parent_key (str) – The key of the feature
• record (tf.Tensor) – String tensor of TF-record

Returns:

A 1-D tensor containing label

Return type:

tf.Tensor