lingvo.tasks.car.evaluation_metadata module

Base class specifying metadata used in evaluation.

class lingvo.tasks.car.evaluation_metadata.EvaluationMetadata(name)

Bases: object

Interface for defining metadata of dataset.

ClassNames()

Returns a list of human-interpretable strings.

LabelMap()

Return a label map of int -> str for each class.

NumClasses()

Total number of classes for the dataset.

DifficultyLevels()

Dictionary of difficulty level strings to int32 indices.

EvalClassIndices()

List of int32 indices for the classes that should be evaled.

IoUThresholds()

Dictionary of IoU thresholds for every evaluated class.

The keys of the dictionary are used to compute EvalClassIndices().

IgnoreClassIndices()

List of int32 indices for the classes that should be ignored.

A detection that matches with a groundtruth bbox of any neighbor class will not be considered as false positive in eval.

NumberOfPrecisionRecallPoints()

Number of points on the precision-recall curve.

MaximumDistance()

Maximum empirically observed Euclidean distance in world coordinates.

DistanceBinWidth()

The width of each bin for Euclidean distance in world coordinates.

MaximumNumberOfPoints()

Maximum empirically observed number of points in bounding box.

NumberOfPointsBins()

Number of logarithmically space bins for examining points.

MaximumRotation()

Maximum rotation angle in world coordinates.

NumberOfRotationBins()

Number of linear spaced rotations to bin.

MinHeight2D()

Minimum height of detections to be evaluated.

Returns

A dictionary of difficulty level strings to ints that are height

thresholds for each level.

RecallAtPrecision()

Report the recall at a given precision level..