Fix get object instances

Author: clinton-encord

encord/objects/ontology_labels_impl.py

@@ -948,8 +948,12 @@ def get_object_instances(
             filtered_frames_list = list()
 
         # Objects in space
+        object_hashes: set[str] = set()  # Needed to filter out duplicate object instances that exist on multiple spaces
         for space in self._space_map.values():
             for object_ in space._objects_map.values():
+                if object_.object_hash in object_hashes:
+                    continue
+
                 # filter by ontology object
                 if not (
                     filter_ontology_object is None
@@ -976,6 +980,7 @@ def get_object_instances(
                             break
 
                 if append:
+                    object_hashes.add(object_.object_hash)
                     ret.append(object_)
 
         # Objects on label row
@@ -1254,8 +1259,13 @@ def get_classification_instances(
             if append:
                 ret.append(classification)
 
+        # Needed to remove filter out duplicate classification instances across spaces
+        classification_hashes: set[str] = set()
         for space in self._space_map.values():
             for classification in space.get_classifications():
+                if classification.classification_hash in classification_hashes:
+                    continue
+
                 if not (
                     filter_ontology_classification is None
                     or classification.ontology_item.feature_node_hash
@@ -1270,11 +1280,24 @@ def get_classification_instances(
                     append = False
 
                 for frame in filtered_frames_list:
-                    if classification.is_on_frame(frame):
-                        append = True
-                        break
+                    if isinstance(space, VideoSpace):
+                        # TODO: In VideoSpace, track classificationHash to ranges to improve performance here.
+                        classifications_on_frames = space._frames_to_classification_hash_to_annotation_data.get(
+                            frame, {}
+                        )
+                        if classification.classification_hash in classifications_on_frames:
+                            append = True
+                            break
+                    elif isinstance(space, ImageSpace):
+                        if frame != 0:
+                            continue
+
+                        if classification.classification_hash in space._classifications_map:
+                            append = True
+                            break
 
                 if append:
+                    classification_hashes.add(classification.classification_hash)
                     ret.append(classification)
 
         return ret

tests/objects/spaces/test_image_space/test_classifications.py

@@ -24,6 +24,8 @@
 checklist_option_1 = checklist_classification.get_child_by_hash("fvLjF0qZ", Option)
 checklist_option_2 = checklist_classification.get_child_by_hash("a4r7nK9i", Option)
 
+radio_classification = all_types_structure.get_child_by_hash("NzIxNTU1", Classification)
+
 
 @pytest.fixture
 def ontology():
@@ -33,6 +35,36 @@ def ontology():
     yield ontology
 
 
+def test_label_row_get_classification_instances_on_image_space(ontology):
+    # Arrange
+    label_row = LabelRowV2(DATA_GROUP_METADATA, Mock(), ontology)
+    label_row.from_labels_dict(DATA_GROUP_TWO_IMAGES_NO_LABELS)
+    image_space_1 = label_row.get_space_by_id("image-1-uuid", type_=ImageSpace)
+    image_space_2 = label_row.get_space_by_id("image-2-uuid", type_=ImageSpace)
+
+    # Act
+    classification_instance_1 = text_classification.create_instance()
+    classification_instance_2 = checklist_classification.create_instance()
+    classification_instance_3 = radio_classification.create_instance()
+
+    # Place classification on space 1
+    image_space_1.place_classification(classification=classification_instance_1)
+    image_space_1.place_classification(classification=classification_instance_2)
+
+    # Place classification on space 2
+    image_space_2.place_classification(classification=classification_instance_3)
+
+    # Assert
+    all_classification_instances = label_row.get_classification_instances()
+    assert len(all_classification_instances) == 3
+
+    classification_instances_on_frame_1 = label_row.get_classification_instances(filter_frames=1)
+    assert len(classification_instances_on_frame_1) == 0
+
+    classification_instances_on_frame_1 = label_row.get_classification_instances(filter_frames=0)
+    assert len(classification_instances_on_frame_1) == 3
+
+
 def test_place_classification_on_image_space(ontology):
     # Arrange
     label_row = LabelRowV2(DATA_GROUP_METADATA, Mock(), ontology)

tests/objects/spaces/test_image_space/test_objects.py

@@ -43,32 +43,22 @@ def test_label_row_get_object_instances_on_space(ontology):
     object_instance_2 = box_ontology_item.create_instance()
 
     # Place objects on space 1
-    image_space_1.place_object(
-        object_instance=object_instance_1,
-        placement=coordinates
-    )
-    image_space_1.place_object(
-        object_instance=object_instance_2,
-        placement=coordinates
-    )
+    image_space_1.place_object(object_instance=object_instance_1, placement=coordinates)
+    image_space_1.place_object(object_instance=object_instance_2, placement=coordinates)
 
     # Place objects on space 2
-    image_space_2.place_object(
-        object_instance=object_instance_1,
-        placement=coordinates
-    )
+    image_space_2.place_object(object_instance=object_instance_1, placement=coordinates)
 
     object_instances = label_row.get_object_instances()
-    assert len(object_instances) == 3
+    assert len(object_instances) == 2
 
     object_instances_on_frame_1 = label_row.get_object_instances(filter_frames=[0])
-    assert len(object_instances_on_frame_1) == 3
+    assert len(object_instances_on_frame_1) == 2
 
     object_instances_on_frame_1 = label_row.get_object_instances(filter_frames=[1])
     assert len(object_instances_on_frame_1) == 0
 
 
-
 def test_place_object_on_image_space(ontology):
     # Arrange
     label_row = LabelRowV2(DATA_GROUP_METADATA, Mock(), ontology)

tests/objects/spaces/test_video_space/test_classifications.py

@@ -20,6 +20,7 @@
 box_with_attributes_ontology_item = all_types_structure.get_child_by_hash("MTA2MjAx", Object)
 box_text_attribute_ontology_item = box_with_attributes_ontology_item.get_child_by_hash("OTkxMjU1", type_=Attribute)
 text_classification = all_types_structure.get_child_by_hash("jPOcEsbw", Classification)
+checklist_classification = all_types_structure.get_child_by_hash("3DuQbFxo", Classification)
 
 
 @pytest.fixture
@@ -30,6 +31,32 @@ def ontology():
     yield ontology
 
 
+def test_label_row_get_classification_instances_on_video_space(ontology):
+    # Arrange
+    label_row = LabelRowV2(DATA_GROUP_METADATA, Mock(), ontology)
+    label_row.from_labels_dict(DATA_GROUP_TWO_VIDEOS_NO_LABELS)
+    video_space_1 = label_row.get_space_by_id("video-1-uuid", type_=VideoSpace)
+    video_space_2 = label_row.get_space_by_id("video-2-uuid", type_=VideoSpace)
+
+    # Act
+    classification_instance_1 = text_classification.create_instance()
+    classification_instance_2 = checklist_classification.create_instance()
+
+    # Place classification on space 1
+    video_space_1.place_classification(classification=classification_instance_1, placement=[0, 1, 2])
+    video_space_1.place_classification(classification=classification_instance_2, placement=[2, 3, 4])
+
+    # Place classification on space 2
+    video_space_2.place_classification(classification=classification_instance_1, placement=[1])
+
+    # Assert
+    all_classification_instances = label_row.get_classification_instances()
+    assert len(all_classification_instances) == 2
+
+    classification_instances_on_frame_1 = label_row.get_classification_instances(filter_frames=1)
+    assert len(classification_instances_on_frame_1) == 1
+
+
 def test_place_classification_on_video_space(ontology):
     # Arrange
     label_row = LabelRowV2(DATA_GROUP_METADATA, Mock(), ontology)

tests/objects/spaces/test_video_space/test_objects.py

@@ -30,6 +30,7 @@ def ontology():
     ontology = Mock(structure=ontology_structure)
     yield ontology
 
+
 def test_label_row_get_object_instances_on_space(ontology):
     label_row = LabelRowV2(DATA_GROUP_METADATA, Mock(), ontology)
     label_row.from_labels_dict(DATA_GROUP_TWO_VIDEOS_NO_LABELS)
@@ -44,30 +45,27 @@ def test_label_row_get_object_instances_on_space(ontology):
     # Place objects on space 1
     video_space_1.place_object(
         object_instance=object_instance_1,
-        placement=FramePlacement(
-            frames=[0, 1, 2], coordinates=coordinates
-        ),
+        placement=FramePlacement(frames=[0, 1, 2], coordinates=coordinates),
     )
     video_space_1.place_object(
         object_instance=object_instance_2,
-        placement=FramePlacement(
-            frames=[2, 3, 4], coordinates=coordinates
-        ),
+        placement=FramePlacement(frames=[2, 3, 4], coordinates=coordinates),
     )
 
     # Place objects on space 2
     video_space_2.place_object(
-        object_instance=object_instance_1,
-        placement=FramePlacement(
-            frames=[1], coordinates=coordinates
-        )
+        object_instance=object_instance_1, placement=FramePlacement(frames=[1], coordinates=coordinates)
     )
 
     object_instances = label_row.get_object_instances()
-    assert len(object_instances) == 3
+    assert len(object_instances) == 2
 
     object_instances_on_frame_1 = label_row.get_object_instances(filter_frames=[1])
-    assert len(object_instances_on_frame_1) == 2
+    assert len(object_instances_on_frame_1) == 1
+
+    object_instances_on_frame_2 = label_row.get_object_instances(filter_frames=[2])
+    assert len(object_instances_on_frame_2) == 2
+
 
 def test_place_object_on_video_space(ontology):
     # Arrange