predictionfilter

predictionfilter ¶

Module for PredictionFilter

Classes¶

PredictionFilter ¶

Bases: ABC

Class to filter predictions

Functions¶

filter `abstractmethod` ¶

filter(prediction_array_xywh)

Filters prediction arrays with corresponding class predictions Args: prediction_array_xywh: input predictions (n x (4 + num_classes))

Returns:

prediction_array_xywh ( ndarray ) –

Filtered predictions (m x (4 + num_classes))

Source code in niceml/utilities/boundingboxes/filtering/predictionfilter.py

@abstractmethod
def filter(self, prediction_array_xywh: np.ndarray) -> np.ndarray:
    """
    Filters prediction arrays with corresponding class predictions
    Args:
        prediction_array_xywh: input predictions (n x (4 + num_classes))

    Returns:
        prediction_array_xywh: Filtered predictions (m x (4 + num_classes))
    """

initialize ¶

initialize(data_description)

Initialization method for the prediction filter Args: data_description: DataDescription that can be used for initialization

Source code in niceml/utilities/boundingboxes/filtering/predictionfilter.py

def initialize(
    self,
    data_description: Union[
        OutputObjDetDataDescription, OutputImageDataDescription
    ],
):
    """
    Initialization method for the prediction filter
    Args:
        data_description: `DataDescription` that can be used for initialization
    """
    if isinstance(data_description, OutputObjDetDataDescription):
        self.output_class_count = data_description.get_output_class_count()
    elif isinstance(data_description, OutputImageDataDescription):
        self.output_class_count = data_description.get_output_channel_count()
    else:
        raise TypeError(
            f"Object of class {type(data_description)} "
            f"is not instance of class: {OutputObjDetDataDescription} "
            f"or {OutputImageDataDescription}"
        )

to_dict ¶

to_dict()

Creates a representation of itself as a dict

Source code in niceml/utilities/boundingboxes/filtering/predictionfilter.py

def to_dict(self) -> dict:
    """Creates a representation of itself as a dict"""
    filter_dict = asdict(self)
    filter_dict["pred_type"] = type(self).__name__
    return filter_dict

Functions¶

get_filter_attributes ¶

get_filter_attributes(filter_class, ignore_attributes=None)

Iterates the attributes of filter_class and returns it. Returns: list of the attributes of filter_class as a dict with name and default as keys

Source code in niceml/utilities/boundingboxes/filtering/predictionfilter.py

def get_filter_attributes(  # QUEST: still used?
    filter_class: Type[PredictionFilter], ignore_attributes: Optional[List[str]] = None
):
    """
    Iterates the attributes of `filter_class` and returns it.
    Returns:
        list of the attributes of `filter_class` as a dict with `name` and `default` as keys
    """
    # pylint: disable=not-an-iterable
    ignore_attributes = ignore_attributes or []
    filter_attributes = []

    for field in fields(filter_class):
        if field.name not in ignore_attributes:
            filter_attribute = {"name": field.name, "default": 0.0, "step": None}

            if field.name in ["score_threshold", "iou_threshold"]:
                filter_attribute["default"] = 0.5

            if field.type == Union[int, None]:
                filter_attribute["step"] = 1
                filter_attribute["default"] = (
                    200 if field.name == "max_output_count" else 0
                )
            filter_attributes.append(filter_attribute)
    return filter_attributes

predictionfilter