Standardize ego_dist and ego_translation

python-sdk/nuscenes/eval/common/data_classes.py

@@ -17,6 +17,7 @@ def __init__(self,
                  size: Tuple[float, float, float] = (0, 0, 0),
                  rotation: Tuple[float, float, float, float] = (0, 0, 0, 0),
                  velocity: Tuple[float, float] = (0, 0),
+                 ego_translation: Tuple[float, float, float] = (0, 0, 0),  # Translation to ego vehicle in meters.
                  num_pts: int = -1):  # Nbr. LIDAR or RADAR inside the box. Only for gt boxes.
 
         # Assert data for shape and NaNs.
@@ -34,6 +35,9 @@ def __init__(self,
         # Velocity can be NaN from our database for certain annotations.
         assert len(velocity) == 2, 'Error: Velocity must have 2 elements!'
 
+        assert len(ego_translation) == 3, 'Error: Translation must have 3 elements!'
+        assert not np.any(np.isnan(ego_translation)), 'Error: Translation may not be NaN!'
+
         assert type(num_pts) == int, 'Error: num_pts must be int!'
         assert not np.any(np.isnan(num_pts)), 'Error: num_pts may not be NaN!'
 
@@ -43,8 +47,14 @@ def __init__(self,
         self.size = size
         self.rotation = rotation
         self.velocity = velocity
+        self.ego_translation = ego_translation
         self.num_pts = num_pts
 
+    @property
+    def ego_dist(self) -> float:
+        """ Compute the distance from this box to the ego vehicle in 2D. """
+        return np.sqrt(np.sum(np.array(self.ego_translation[:2]) ** 2))
+
     def __repr__(self):
         return str(self.serialize())
 

python-sdk/nuscenes/eval/common/loaders.py

@@ -195,9 +195,7 @@ def add_center_dist(nusc: NuScenes,
             ego_translation = (box.translation[0] - pose_record['translation'][0],
                                box.translation[1] - pose_record['translation'][1],
                                box.translation[2] - pose_record['translation'][2])
-            if isinstance(box, DetectionBox):
-                box.ego_dist = np.sqrt(np.sum(np.array(ego_translation[:2]) ** 2))
-            elif isinstance(box, TrackingBox):
+            if isinstance(box, DetectionBox) or isinstance(box, TrackingBox):
                 box.ego_translation = ego_translation
             else:
                 raise NotImplementedError

python-sdk/nuscenes/eval/detection/algo.py

@@ -63,9 +63,7 @@ def accumulate(gt_boxes: EvalBoxes,
                   'scale_err': [],
                   'orient_err': [],
                   'attr_err': [],
-                  'conf': [],
-                  'ego_dist': [],
-                  'vel_magn': []}
+                  'conf': []}
 
     # ---------------------------------------------
     # Match and accumulate match data.
@@ -111,10 +109,6 @@ def accumulate(gt_boxes: EvalBoxes,
             match_data['attr_err'].append(1 - attr_acc(gt_box_match, pred_box))
             match_data['conf'].append(pred_box.detection_score)
 
-            # For debugging only.
-            match_data['ego_dist'].append(gt_box_match.ego_dist)
-            match_data['vel_magn'].append(np.sqrt(np.sum(np.array(gt_box_match.velocity) ** 2)))
-
         else:
             # No match. Mark this as a false positive.
             tp.append(0)

python-sdk/nuscenes/eval/detection/data_classes.py

@@ -321,13 +321,13 @@ def __init__(self,
                  size: Tuple[float, float, float] = (0, 0, 0),
                  rotation: Tuple[float, float, float, float] = (0, 0, 0, 0),
                  velocity: Tuple[float, float] = (0, 0),
-                 ego_dist: float = 0.0,  # Distance to ego vehicle in meters.
+                 ego_translation: [float, float, float] = (0, 0, 0),  # Translation to ego vehicle in meters.
                  num_pts: int = -1,  # Nbr. LIDAR or RADAR inside the box. Only for gt boxes.
                  detection_name: str = 'car',  # The class name used in the detection challenge.
                  detection_score: float = -1.0,  # GT samples do not have a score.
                  attribute_name: str = ''):  # Box attribute. Each box can have at most 1 attribute.
 
-        super().__init__(sample_token, translation, size, rotation, velocity, num_pts)
+        super().__init__(sample_token, translation, size, rotation, velocity, ego_translation, num_pts)
 
         assert detection_name is not None, 'Error: detection_name cannot be empty!'
         assert detection_name in DETECTION_NAMES, 'Error: Unknown detection_name %s' % detection_name
@@ -339,7 +339,6 @@ def __init__(self,
         assert not np.any(np.isnan(detection_score)), 'Error: detection_score may not be NaN!'
 
         # Assign.
-        self.ego_dist = ego_dist
         self.detection_name = detection_name
         self.detection_score = detection_score
         self.attribute_name = attribute_name
@@ -350,7 +349,7 @@ def __eq__(self, other):
                 self.size == other.size and
                 self.rotation == other.rotation and
                 self.velocity == other.velocity and
-                self.ego_dist == other.ego_dist and
+                self.ego_translation == other.ego_translation and
                 self.num_pts == other.num_pts and
                 self.detection_name == other.detection_name and
                 self.detection_score == other.detection_score and
@@ -364,7 +363,7 @@ def serialize(self) -> dict:
             'size': self.size,
             'rotation': self.rotation,
             'velocity': self.velocity,
-            'ego_dist': self.ego_dist,
+            'ego_translation': self.ego_translation,
             'num_pts': self.num_pts,
             'detection_name': self.detection_name,
             'detection_score': self.detection_score,
@@ -379,7 +378,8 @@ def deserialize(cls, content: dict):
                    size=tuple(content['size']),
                    rotation=tuple(content['rotation']),
                    velocity=tuple(content['velocity']),
-                   ego_dist=0.0 if 'ego_dist' not in content else float(content['ego_dist']),
+                   ego_translation=(0.0, 0.0, 0.0) if 'ego_translation' not in content
+                   else tuple(content['ego_translation']),
                    num_pts=-1 if 'num_pts' not in content else int(content['num_pts']),
                    detection_name=content['detection_name'],
                    detection_score=-1.0 if 'detection_score' not in content else float(content['detection_score']),

python-sdk/nuscenes/eval/detection/tests/test_algo.py

@@ -58,7 +58,7 @@ def random_attr():
                     detection_name=detection_name,
                     detection_score=random.random(),
                     attribute_name=random_attr(),
-                    ego_dist=random.random()*10,
+                    ego_translation=(random.random() * 10, 0, 0),
                 ))
             gt.add_boxes(str(sample_itt), this_gt)
 
@@ -76,7 +76,7 @@ def random_attr():
                     detection_name=detection_name,
                     detection_score=random.random(),
                     attribute_name=random_attr(),
-                    ego_dist=random.random() * 10,
+                    ego_translation=(random.random() * 10, 0, 0),
                 ))
 
             pred.add_boxes(str(sample_itt), this_pred)

python-sdk/nuscenes/eval/detection/tests/test_loader.py

@@ -88,12 +88,13 @@ def test_filter_eval_boxes(self):
         self.assertEqual(filtered_boxes.boxes[sample_token][1].detection_name, 'bicycle')
         self.assertEqual(filtered_boxes.boxes[sample_token][1].translation[0], 68.681)
 
-        # Add another bike on the bike rack center but set the ego_dist higher than what's defined in max_dist
+        # Add another bike on the bike rack center,
+        # but set the ego_dist (derived from ego_translation) higher than what's defined in max_dist
         box5 = DetectionBox(sample_token=sample_token,
                             translation=(683.681, 1592.002, 0.809),
                             size=(1, 1, 1),
                             detection_name='bicycle',
-                            ego_dist=100.0)
+                            ego_translation=(100.0, 0.0, 0.0))
 
         eval_boxes = EvalBoxes()
         eval_boxes.add_boxes(sample_token, [box1, box2, box3, box4, box5])
@@ -127,25 +128,25 @@ def test_filter_eval_boxes(self):
                             translation=(683.681, 1592.002, 0.809),
                             size=(1, 1, 1),
                             detection_name='bicycle',
-                            ego_dist=25.0)
+                            ego_translation=(25.0, 0.0, 0.0))
 
         box2 = DetectionBox(sample_token=sample_token,
                             translation=(683.681, 1592.002, 0.809),
                             size=(1, 1, 1),
                             detection_name='motorcycle',
-                            ego_dist=45.0)
+                            ego_translation=(45.0, 0.0, 0.0))
 
         box3 = DetectionBox(sample_token=sample_token,
                             translation=(683.681, 1592.002, 0.809),
                             size=(1, 1, 1),
                             detection_name='car',
-                            ego_dist=45.0)
+                            ego_translation=(45.0, 0.0, 0.0))
 
         box4 = DetectionBox(sample_token=sample_token,
                             translation=(683.681, 1592.002, 0.809),
                             size=(1, 1, 1),
                             detection_name='car',
-                            ego_dist=55.0)
+                            ego_translation=(55.0, 0.0, 0.0))
 
         box5 = DetectionBox(sample_token=sample_token,
                             translation=(683.681, 1592.002, 0.809),

python-sdk/nuscenes/eval/tracking/data_classes.py

@@ -258,16 +258,13 @@ def __init__(self,
                  size: Tuple[float, float, float] = (0, 0, 0),
                  rotation: Tuple[float, float, float, float] = (0, 0, 0, 0),
                  velocity: Tuple[float, float] = (0, 0),
-                 ego_translation: [float, float, float] = (0, 0, 0),  # Translation to ego vehicle in meters.
+                 ego_translation: Tuple[float, float, float] = (0, 0, 0),  # Translation to ego vehicle in meters.
                  num_pts: int = -1,  # Nbr. LIDAR or RADAR inside the box. Only for gt boxes.
                  tracking_id: str = '',  # Instance id of this object.
                  tracking_name: str = '',  # The class name used in the tracking challenge.
                  tracking_score: float = -1.0):  # Does not apply to GT.
 
-        super().__init__(sample_token, translation, size, rotation, velocity, num_pts)
-
-        assert len(ego_translation) == 3, 'Error: Translation must have 3 elements!'
-        assert not np.any(np.isnan(ego_translation)), 'Error: Translation may not be NaN!'
+        super().__init__(sample_token, translation, size, rotation, velocity, ego_translation, num_pts)
 
         assert tracking_name is not None, 'Error: tracking_name cannot be empty!'
         assert tracking_name in TRACKING_NAMES, 'Error: Unknown tracking_name %s' % tracking_name
@@ -276,16 +273,10 @@ def __init__(self,
         assert not np.any(np.isnan(tracking_score)), 'Error: tracking_score may not be NaN!'
 
         # Assign.
-        self.ego_translation = ego_translation
         self.tracking_id = tracking_id
         self.tracking_name = tracking_name
         self.tracking_score = tracking_score
 
-    @ property
-    def ego_dist(self) -> float:
-        """ Compute the distance from this box to the ego vehicle in 2D. """
-        return np.sqrt(np.sum(np.array(self.ego_translation[:2]) ** 2))
-
     def __eq__(self, other):
         return (self.sample_token == other.sample_token and
                 self.translation == other.translation and