def _evaluate_single(self, detections_dict, groundtruth_list):
groundtruth_dict = {
'annotations': groundtruth_list,
'images': [{
'id': gt['image_id']
} for gt in groundtruth_list],
'categories': self.categories
}
if len(groundtruth_list) > 0:
detections_list = coco_tools.ExportSingleImageDetectionBoxesToCoco(
image_id=groundtruth_list[0]['image_id'],
category_id_set=set([c['id'] for c in self.categories]),
detection_boxes=detections_dict['detection_boxes'],
detection_scores=detections_dict['detection_scores'],
detection_classes=detections_dict['detection_classes'])
else:
detections_list = []
# The COCO evaluation prints some information, which we don't care about
with HiddenPrints():
groundtruth = coco_tools.COCOWrapper(groundtruth_dict)
detections = groundtruth.LoadAnnotations(detections_list)
evaluator = coco_tools.COCOEvalWrapper(groundtruth,
detections,
iou_type='bbox')
summary_metrics, _ = evaluator.ComputeMetrics()
return summary_metrics['Precision/mAP']
def add_single_detected_image_info(self, image_id, detections_dict):
"""Add detection results of all frames to the eval pipeline.
This method overrides the function defined in the base class.
Args:
image_id: A unique string/integer identifier for the image.
detections_dict: A list of dictionary containing -
DetectionResultFields.detection_boxes: float32 numpy array of shape
[num_boxes, 4] containing `num_boxes` detection boxes of the format
[ymin, xmin, ymax, xmax] in absolute image coordinates.
DetectionResultFields.detection_scores: float32 numpy array of shape
[num_boxes] containing detection scores for the boxes.
DetectionResultFields.detection_classes: integer numpy array of shape
[num_boxes] containing 1-indexed detection classes for the boxes.
Raises:
ValueError: If groundtruth for the image_id is not available.
"""
for idx, det in enumerate(detections_dict):
if not det:
continue
image_frame_id = '{}_{}'.format(image_id, idx)
if image_frame_id not in self._image_ids:
raise ValueError(
'Missing groundtruth for image-frame id: {}'.format(
image_frame_id))
if self._image_ids[image_frame_id]:
tf.logging.warning(
'Ignoring detection with image id %s since it was '
'previously added', image_frame_id)
continue
self._detection_boxes_list.extend(
coco_tools.ExportSingleImageDetectionBoxesToCoco(
image_id=image_frame_id,
category_id_set=self._category_id_set,
detection_boxes=det[
standard_fields.DetectionResultFields.detection_boxes],
detection_scores=det[standard_fields.DetectionResultFields.
detection_scores],
detection_classes=det[standard_fields.DetectionResultFields
.detection_classes]))
self._image_ids[image_frame_id] = True
def add_single_detected_image_info(self,
image_id,
detections_dict):
"""Adds detections for a single image to be used for evaluation.
If a detection has already been added for this image id, a warning is
logged, and the detection is skipped.
Args:
image_id: A unique string/integer identifier for the image.
detections_dict: A dictionary containing -
DetectionResultFields.detection_boxes: float32 numpy array of shape
[num_boxes, 4] containing `num_boxes` detection boxes of the format
[ymin, xmin, ymax, xmax] in absolute image coordinates.
DetectionResultFields.detection_scores: float32 numpy array of shape
[num_boxes] containing detection scores for the boxes.
DetectionResultFields.detection_classes: integer numpy array of shape
[num_boxes] containing 1-indexed detection classes for the boxes.
Raises:
ValueError: If groundtruth for the image_id is not available.
"""
if image_id not in self._image_ids:
raise ValueError('Missing groundtruth for image id: {}'.format(image_id))
if self._image_ids[image_id]:
tf.logging.warning('Ignoring detection with image id %s since it was '
'previously added', image_id)
return
self._detection_boxes_list.extend(
coco_tools.ExportSingleImageDetectionBoxesToCoco(
image_id=image_id,
category_id_set=self._category_id_set,
detection_boxes=detections_dict[standard_fields.
DetectionResultFields
.detection_boxes],
detection_scores=detections_dict[standard_fields.
DetectionResultFields.
detection_scores],
detection_classes=detections_dict[standard_fields.
DetectionResultFields.
detection_classes]))
self._image_ids[image_id] = True
def testSingleImageDetectionBoxesExportWithKeypoints(self):
boxes = np.array([[0, 0, 1, 1], [0, 0, .5, .5], [.5, .5, 1, 1]],
dtype=np.float32)
coco_boxes = np.array([[0, 0, 1, 1], [0, 0, .5, .5], [.5, .5, .5, .5]],
dtype=np.float32)
keypoints = np.array([[[0, 0], [0.25, 0.25], [0.75, 0.75]],
[[0, 0], [0.125, 0.125], [0.375, 0.375]],
[[0.5, 0.5], [0.75, 0.75], [1.0, 1.0]]],
dtype=np.float32)
visibilities = np.array([[2, 2, 2], [2, 2, 2], [2, 2, 2]],
dtype=np.int32)
classes = np.array([1, 2, 3], dtype=np.int32)
scores = np.array([0.8, 0.2, 0.7], dtype=np.float32)
# Tests exporting without passing in is_crowd (for backward compatibility).
coco_annotations = coco_tools.ExportSingleImageDetectionBoxesToCoco(
image_id='first_image',
category_id_set=set([1, 2, 3]),
detection_boxes=boxes,
detection_scores=scores,
detection_classes=classes,
detection_keypoints=keypoints,
detection_keypoint_visibilities=visibilities)
for i, annotation in enumerate(coco_annotations):
self.assertTrue(
np.all(np.isclose(annotation['bbox'], coco_boxes[i])))
self.assertEqual(annotation['image_id'], 'first_image')
self.assertEqual(annotation['category_id'], classes[i])
self.assertTrue(
np.all(np.isclose(annotation['bbox'], coco_boxes[i])))
self.assertEqual(annotation['score'], scores[i])
self.assertEqual(annotation['num_keypoints'], 3)
self.assertTrue(
np.all(
np.isclose(annotation['keypoints'][0::3], keypoints[i, :,
1])))
self.assertTrue(
np.all(
np.isclose(annotation['keypoints'][1::3], keypoints[i, :,
0])))
self.assertTrue(
np.all(np.equal(annotation['keypoints'][2::3],
visibilities[i])))
def testSingleImageDetectionBoxesExport(self):
boxes = np.array([[0, 0, 1, 1], [0, 0, .5, .5], [.5, .5, 1, 1]],
dtype=np.float32)
classes = np.array([1, 2, 3], dtype=np.int32)
scores = np.array([0.8, 0.2, 0.7], dtype=np.float32)
coco_boxes = np.array([[0, 0, 1, 1], [0, 0, .5, .5], [.5, .5, .5, .5]],
dtype=np.float32)
coco_annotations = coco_tools.ExportSingleImageDetectionBoxesToCoco(
image_id='first_image',
category_id_set=set([1, 2, 3]),
detection_boxes=boxes,
detection_classes=classes,
detection_scores=scores)
for i, annotation in enumerate(coco_annotations):
self.assertEqual(annotation['image_id'], 'first_image')
self.assertEqual(annotation['category_id'], classes[i])
self.assertAlmostEqual(annotation['score'], scores[i])
self.assertTrue(
np.all(np.isclose(annotation['bbox'], coco_boxes[i])))
