compute_matches#

DetectionEvaluator.compute_matches(predictions_names: str | Iterable[str] | None = None, min_iou: float = 0, category_agnostic: bool = False) → dict[str, DataFrame][source]#

Get matches between predictions and targets of the Evaluator.

Related tutorial

Parameters:

predictions_names – name or collection of prediction names to compute the matches on. If set to None, will compute the matches with the prediction DataFrames contained in the self.predictions_dictionary attribute. Defaults to None
min_iou – IoU above which the detection is considered valid. Defaults to 0. Note that the lower bound of min_iou is not inclusive.
category_agnostic – if set to False, matches are computed between categories, otherwise, matches are computed globally

Returns:

dict of DataFrame of matches, one entry per prediction specified in arguments. Will contain prediction_id and groundtruth_id columns. Index is unrelevant. Each prediction id and target id should appear once and only once. As such, at worse (no match at all), the dataframe will be \(N+M\) rows with \(N\) the number of predictions and \(M\) the number of targets, and at best it will be \(max(M,N)\)

Example

>>> from lours.utils.doc_utils import dummy_dataset
>>> groundtruth = dummy_dataset(
...     10,
...     1000,
...     label_map={0: "person", 1: "car"},
... )
>>> predictions1 = dummy_dataset(
...     10,
...     10000,
...     label_map=groundtruth.label_map,
...     images=groundtruth.images,
...     add_confidence=True,
...     seed=0,
... )
>>> predictions2 = dummy_dataset(
...     10,
...     10000,
...     label_map=groundtruth.label_map,
...     images=groundtruth.images,
...     add_confidence=True,
...     seed=1,
... )
>>> evaluator = DetectionEvaluator(
...     groundtruth=groundtruth, A=predictions1, B=predictions2
... )
>>> matches = evaluator.compute_matches()
computing matches between groundtruth and A (category specific)
computing matches between groundtruth and B (category specific)
>>> matches["A"]
      prediction_id       iou  groundtruth_id
0             2311  0.370857             207
1              515  0.586261             820
2             7071  0.468022             585
3             4444  0.089832              87
4              235  0.431787             105
..             ...       ...             ...
487           5016  0.000000            <NA>
488           3608  0.000000            <NA>
489            437  0.000000            <NA>
490           8837  0.000000            <NA>
491           2508  0.000000            <NA>

[10000 rows x 3 columns]

You can select a particular set of prediction to only compute them

>>> B_matches = evaluator.compute_matches(
...     predictions_names="B", category_agnostic=True
... )
computing matches between groundtruth and B (category agnostic)
>>> B_matches["B"]
       prediction_id       iou  groundtruth_id
0               7849  0.267152             832
1               8819  0.089308             130
2               6537  0.322729             785
3               1616  0.406822             326
4               8021  0.510778             929
...              ...       ...             ...
1022            7377  0.000000            <NA>
1023            8370  0.000000            <NA>
1024            3534  0.000000            <NA>
1025            7087  0.000000            <NA>
1026            1410  0.000000            <NA>

[10000 rows x 3 columns]