MetaPropertiesExtractor

The MetaPropertiesExtractor class allows the automatic extraction of meta-annotations from the observations.
Odin provides the support to the extraction of two different meta-annotations:

MetaProperties.COLOR: extraction of the image color space. It can be ‘rgb’ (colored) or ‘bw’ (greyscaled).
MetaProperties.FACES: extraction of the number of faces present in the image. It can be ‘0-1’, ‘2-4’, ‘5+’.

Parameters

dataset: DatasetClassification; Dataset used to perform the meta-annotations extraction.
properties: MetaProperties, optional; List of the meta-annotations to be included in the extraction. If None, no default meta-annotation will be extracted.
(default is None)
output_path: str, optional; Path used to save the new generated annotated data set. If not provided, it is saved in the same location of the old data set.
(default is None)

Example

from odin.annotator.meta_annotator_extractor import MetaPropertiesExtractor, MetaProperties

# define the properties to be extracted
my_properties = [MetaProperties.COLOR, MetaProperties.FACES]
my_output_path = "path/to/save/the/new/annotated/dataset/"

my_annotator = MetaPropertiesExtractor(my_classification_dataset,
                                       properties=my_properties,
                                       output_path=my_output_path)
my_annotator.start_annotation()

Custom extractor

It is possible to implement a custom meta-annotation extractor by simply implement the process_object function of the PropertyAnnotatorInterface interface. The following example shows how to implement a custom extractor and how it can be integrated into the MetaPropertiesExtractor.

Example

from odin.annotator.meta_annotator_extractor import MetaPropertiesExtractor, MetaProperties, PropertyAnnotatorInterface

# define your custom meta-annotation extractor
class MyCustomMetaAnnotationExtractor(PropertyAnnotatorInterface):
    DEFAULT_VALUE = 0
    NAME = "n_categories"

    def __init__(self):
        property_values = []
        super().__init__(self.NAME, property_values, self.DEFAULT_VALUE)

    def process_object(self, data_object, dataset_abs_path=None):
        """
        Return the number of categories in the observation
        """
        n_categories = len(data_object["categories"])
        if n_categories not in self.property_values:
            self.property_values.append(n_categories)
        return n_categories

# instantiate the MetaPropertiesExtractor
my_output_path = "path/to/save/the/new/annotated/dataset/"
my_annotator = MetaPropertiesExtractor(my_classification_dataset,
                                       output_path=my_output_path)

# instantiate the custom extractor and add it to the annotator
custom_extractor = MyCustomMetaAnnotationExtractor()
my_annotator.add_custom_property(custom_extractor)

my_annotator.start_annotation()

Tasks supported

Binary Classification	Single-label Classification	Multi-label Classification	Object Detection	Instance Segmentation
yes	yes	yes	no	no