CrowdDetectionEvaluator#

class CrowdDetectionEvaluator(groundtruth: Dataset, name: str | None = None, **predictions: Dataset)[source]#

Bases: DetectionEvaluatorBase

Class specialization for crowd detection and counting tasks. Note that the constructor is the same as the base Evaluator

Constructor of the Evaluator object.

Parameters:

groundtruth – Dataset object representing the ground truth with annotations, image data and label_map
name – Name of Evaluator. If set to None, will be deduced from groundtruth’s dataset name
**predictions – keyword arguments for additional datasets to compare the groundtruth to. Its images must match the groundtruth dataset (see add_prediction_dataset method below).

Attributes

name: str | None#: Name of Evaluator. Can be deduced from groundtruth’s dataset name and will be used in export functions like DetectionEvaluator.to_fiftyone()

groundtruth: pd.DataFrame#: DataFrame comprising annotation data. Must have at least image_id column

predictions_dictionary: dict[str, pd.DataFrame]#: dictionary of DataFrames comprising prediction data. Must have at least image_id and confidence columns

images: pd.DataFrame#: DataFrame comprising image data. This dataframe should be referred to by both gt and predictions with the image_id column

images_root: Path#: Root folder where to grab images. Image filepath will be concatenation of images_root and their relative path

label_map: dict[int, str]#: Mapping from category_id to category_str. If used, is generally taken from the groundtruth Dataset. The prediction must be compatible with it

Methods

`add_predictions`(predictions_name, predictions)	Method to add predictions to the Evaluator from a dataframe.
`add_predictions_dataset`(predictions_name, ...)	Method to add predictions to the Evaluator from a Dataset object.
`compute_count_error`([groups, quantiles, ...])	Compute Count error metrics, both absolute (in number of objects found) and relative (with respect to groundtruth number of objects) with respect to confidence threshold.
`compute_normalized_precision_recall`()	Compute nAP between detected points and ground truth according to the algorithm proposed in [Ref]
`from_parquet`(input_dir)	Class method to construct an instance of this class or a subclass.
`get_annotations_attributes`([predictions_name])	Get the name of columns related to annotations attributes.
`get_image_attributes`()	Get the name of columns related to image attributes.
`to_fiftyone`([name, record_fo_ids, existing])	Convert evaluator to fiftyone.
`to_parquet`(output_dir[, overwrite])	Save the current object to a folder containing parquet files for dataframes inside this object, and a metadata.yaml file for other attributes.