iaa_od.utils.fatigue_evaluation

Functions

`extract_fatigue_data`(result, timing_files)	Extracts fatigue data from the provided timing files and computes annotation sessions for each Ground Truth (GT).
`per_image_fatigue_granularities`(result)	Compute per-image count and size granularities for the whole dataset, normalised against the dataset-wide benchmark so the values are comparable across images.

iaa_od.utils.fatigue_evaluation.extract_fatigue_data(result, timing_files)[source]

Extracts fatigue data from the provided timing files and computes annotation sessions for each Ground Truth (GT).

Parameters:

result (Result) – The Result object containing the Ground Truths and Units.
timing_files (list[str]) – A list of strings representing the paths to the JSON timing files

Returns:

A dictionary where the keys are GT names (strings) and the values are dictionaries mapping session IDs (integers) to lists of AnnotationTimestamp objects.

Return type:

dict[str, dict[int, list]]

iaa_od.utils.fatigue_evaluation.per_image_fatigue_granularities(result)[source]

Compute per-image count and size granularities for the whole dataset, normalised against the dataset-wide benchmark so the values are comparable across images.

This is the granularity counterpart to result.alpha_per_image: it is what show_annotator_fatigue_granularity plots against the normalised timestamps. The dataset-wide benchmark (rather than each image’s own median) is essential here, since a temporal view is only meaningful if a value of, say, 1.4 means the same thing on the first image of a session as it does on the last.

The returned mapping is keyed by image filename, matching AnnotationTimestamp.image_filename. Images whose units are all singletons, and annotators that did not contribute to any non-singleton unit on a given image, are simply absent from the corresponding entry; the plotting code treats those as missing (NaN) data points.

Parameters:: result (Result) – The Result object containing the Ground Truths and Units.
Returns:: A mapping from image filename to that image’s Granularities.
Return type:: dict[str, Granularities]