def test_classification_per_class_accuracy_prediction_top3(self):
annotation = [
ClassificationAnnotation('identifier_1', 1),
ClassificationAnnotation('identifier_2', 1)
]
prediction = [
ClassificationPrediction('identifier_1', [1.0, 2.0, 3.0, 4.0]),
ClassificationPrediction('identifier_2', [2.0, 1.0, 3.0, 4.0])
]
dataset = DummyDataset(label_map={0: '0', 1: '1', 2: '2', 3: '3'})
dispatcher = MetricsExecutor([{
'type': 'accuracy_per_class',
'top_k': 3
}], dataset)
dispatcher.update_metrics_on_batch(range(len(annotation)), annotation,
prediction)
for _, evaluation_result in dispatcher.iterate_metrics(
annotation, prediction):
assert evaluation_result.name == 'accuracy_per_class'
assert len(evaluation_result.evaluated_value) == 4
assert evaluation_result.evaluated_value[0] == pytest.approx(0.0)
assert evaluation_result.evaluated_value[1] == pytest.approx(0.5)
assert evaluation_result.evaluated_value[2] == pytest.approx(0.0)
assert evaluation_result.evaluated_value[3] == pytest.approx(0.0)
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_classification_accuracy_result_for_batch_2_with_not_ordered_ids(
self):
annotations = [
ClassificationAnnotation('identifier', 3),
ClassificationAnnotation('identifier1', 1)
]
predictions = [
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0]),
ClassificationPrediction('identifier2', [1.0, 1.0, 1.0, 4.0])
]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 1}], None)
metric_result, _ = dispatcher.update_metrics_on_batch([42, 17],
annotations,
predictions)
expected_metric_result = [
PerImageMetricResult('accuracy', 'accuracy', 1.0, 'higher-better'),
PerImageMetricResult('accuracy', 'accuracy', 0.0, 'higher-better')
]
assert len(metric_result) == 2
assert 42 in metric_result
assert len(metric_result[42]) == 1
assert metric_result[42][0] == expected_metric_result[0]
assert 17 in metric_result
assert len(metric_result[17]) == 1
assert metric_result[17][0] == expected_metric_result[1]
def test_classification_accuracy_result_for_batch_1_with_2_metrics(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
]
dispatcher = MetricsExecutor([{
'name': 'top1',
'type': 'accuracy',
'top_k': 1
}, {
'name': 'top3',
'type': 'accuracy',
'top_k': 3
}], None)
metric_result, _ = dispatcher.update_metrics_on_batch(
range(len(annotations)), annotations, predictions)
expected_metric_result = [
PerImageMetricResult('top1', 'accuracy', 1.0, 'higher-better'),
PerImageMetricResult('top3', 'accuracy', 1.0, 'higher-better')
]
assert len(metric_result) == 1
assert 0 in metric_result
assert len(metric_result[0]) == 2
assert metric_result[0][0] == expected_metric_result[0]
assert metric_result[0][1] == expected_metric_result[1]
def test_complete_accuracy_with_container_sources(self):
annotations = [
ContainerAnnotation(
{'a': ClassificationAnnotation('identifier', 3)})
]
predictions = [
ContainerPrediction({
'p':
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
})
]
config = [{
'type': 'accuracy',
'top_k': 1,
'annotation_source': 'a',
'prediction_source': 'p'
}]
dispatcher = MetricsExecutor(config, None)
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
for _, evaluation_result in dispatcher.iterate_metrics(
annotations, predictions):
assert evaluation_result.name == 'accuracy'
assert evaluation_result.evaluated_value == pytest.approx(1.0)
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_accuracy_on_prediction_container_with_several_suitable_representations_raise_config_error_exception(
self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ContainerPrediction({
'prediction1':
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0]),
'prediction2':
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
})
]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 1}], None)
with pytest.raises(ConfigError):
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
def test_accuracy_with_wrong_annotation_type_raise_config_error_exception(
self):
annotations = [DetectionAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 1}], None)
with pytest.raises(ConfigError):
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
def test_config_default_presenter(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
]
config = [{'type': 'accuracy', 'top_k': 1}]
dispatcher = MetricsExecutor(config, None)
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
for presenter, _ in dispatcher.iterate_metrics(annotations,
predictions):
assert isinstance(presenter, ScalarPrintPresenter)
def test_zero_accuracy_top_3(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [5.0, 3.0, 4.0, 1.0])
]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 3}], None)
for _, evaluation_result in dispatcher.iterate_metrics(
annotations, predictions):
assert evaluation_result.name == 'accuracy'
assert evaluation_result.evaluated_value == 0.0
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_complete_accuracy_top_3(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [1.0, 3.0, 4.0, 2.0])
]
dispatcher = MetricsExecutor([{'type': 'accuracy', 'top_k': 3}], None)
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
for _, evaluation_result in dispatcher.iterate_metrics(
annotations, predictions):
assert evaluation_result.name == 'accuracy'
assert evaluation_result.evaluated_value == pytest.approx(1.0)
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_threshold_is_10_by_config(self):
annotations = [ClassificationAnnotation('identifier', 3)]
predictions = [
ClassificationPrediction('identifier', [5.0, 3.0, 4.0, 1.0])
]
dispatcher = MetricsExecutor([{
'type': 'accuracy',
'top_k': 3,
'abs_threshold': 10
}], None)
for _, evaluation_result in dispatcher.iterate_metrics([annotations],
[predictions]):
assert evaluation_result.name == 'accuracy'
assert evaluation_result.evaluated_value == 0.0
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold == 10
def test_accuracy_with_unsupported_annotation_type_as_annotation_source_for_container_raises_config_error(
self):
annotations = [
ContainerAnnotation(
{'annotation': DetectionAnnotation('identifier', 3)})
]
predictions = [
ClassificationPrediction('identifier', [1.0, 1.0, 1.0, 4.0])
]
dispatcher = MetricsExecutor([{
'type': 'accuracy',
'top_k': 1,
'annotation_source': 'annotation'
}], None)
with pytest.raises(ConfigError):
dispatcher.update_metrics_on_batch(range(len(annotations)),
annotations, predictions)
def test_classification_per_class_accuracy_fully_zero_prediction(self):
annotation = ClassificationAnnotation('identifier', 0)
prediction = ClassificationPrediction('identifier', [1.0, 2.0])
dataset = DummyDataset(label_map={0: '0', 1: '1'})
dispatcher = MetricsExecutor([{
'type': 'accuracy_per_class',
'top_k': 1
}], dataset)
dispatcher.update_metrics_on_batch(range(1), [annotation],
[prediction])
for _, evaluation_result in dispatcher.iterate_metrics([annotation],
[prediction]):
assert evaluation_result.name == 'accuracy_per_class'
assert len(evaluation_result.evaluated_value) == 2
assert evaluation_result.evaluated_value[0] == pytest.approx(0.0)
assert evaluation_result.evaluated_value[1] == pytest.approx(0.0)
assert evaluation_result.reference_value is None
assert evaluation_result.abs_threshold is None
assert evaluation_result.rel_threshold is None
def test_predictions_loading_with_adapter(self, mocker):
launcher_config = {
'framework': 'dummy',
'loader': 'pickle',
'data_path': '/path'
}
raw_prediction_batch = StoredPredictionBatch(
{'prediction': np.array([[0, 1]])}, [1], [{}])
expected_prediction = ClassificationPrediction(1, np.array([0, 1]))
adapter = ClassificationAdapter({'type': 'classification'})
mocker.patch(
'openvino.tools.accuracy_checker.launcher.loaders.pickle_loader.PickleLoader.read_pickle',
return_value=[raw_prediction_batch])
launcher = DummyLauncher(launcher_config, adapter=adapter)
assert len(launcher._loader.data) == 1
prediction = launcher.predict([1])
assert len(prediction) == 1
assert isinstance(prediction[0], ClassificationPrediction)
assert prediction[0].identifier == expected_prediction.identifier
assert np.array_equal(prediction[0].scores, expected_prediction.scores)