resolve converter issues

Author: Matt Sokoloff

labelbox/data/annotation_types/annotation.py

@@ -1,7 +1,5 @@
 from typing import Any, Dict, List, Union
 
-from pydantic.main import BaseModel
-
 from .classification import Checklist, Dropdown, Radio, Text
 from .feature import FeatureSchema
 from .geometry import Geometry

labelbox/data/annotation_types/classification/classification.py

@@ -1,10 +1,24 @@
 from typing import Any, Dict, List
 
-from pydantic.main import BaseModel
+try:
+    from typing import Literal
+except:
+    from typing_extensions import Literal
 
+from pydantic import BaseModel, validator
 from ..feature import FeatureSchema
 
 
+# TODO: Replace when pydantic adds support for unions that don't coerce types
+class _TempName(BaseModel):
+    name: str
+
+    def dict(self, *args, **kwargs):
+        res = super().dict(*args, **kwargs)
+        res.pop('name')
+        return res
+
+
 class ClassificationAnswer(FeatureSchema):
     """
     - Represents a classification option.
@@ -19,8 +33,9 @@ class Radio(BaseModel):
     answer: ClassificationAnswer
 
 
-class Checklist(BaseModel):
+class Checklist(_TempName):
     """ A classification with many selected options allowed """
+    name: Literal["checklist"] = "checklist"
     answer: List[ClassificationAnswer]
 
 
@@ -29,9 +44,10 @@ class Text(BaseModel):
     answer: str
 
 
-class Dropdown(BaseModel):
+class Dropdown(_TempName):
     """
     - A classification with many selected options allowed .
     - This is not currently compatible with MAL.
     """
+    name: Literal["dropdown"] = "dropdown"
     answer: List[ClassificationAnswer]

labelbox/data/serialization/labelbox_v1/classification.py

@@ -82,7 +82,7 @@ def to_common(self) -> Dropdown:
     def from_common(cls, dropdown: Dropdown, feature_schema_id: Cuid,
                     **extra) -> "LBV1Dropdown":
         return cls(schema_id=feature_schema_id,
-                   answers=[
+                   answer=[
                        LBV1ClassificationAnswer(
                            schema_id=answer.feature_schema_id,
                            title=answer.name,

labelbox/data/serialization/labelbox_v1/converter.py

@@ -45,6 +45,7 @@ def deserialize(json_data: Iterable[Dict[str, Any]]) -> LabelGenerator:
         Returns:
             LabelGenerator containing the export data.
         """
+
         def label_generator():
             for example in json_data:
                 if 'frames' in example['Label']:
@@ -54,6 +55,7 @@ def label_generator():
                 if example['Label']:
                     # Don't construct empty dict
                     yield LBV1Label(**example).to_common()
+
         return LabelGenerator(data=label_generator())
 
     @staticmethod

labelbox/data/serialization/labelbox_v1/objects.py

@@ -15,7 +15,9 @@
 class LBV1ObjectBase(LBV1Feature):
     color: Optional[str] = None
     instanceURI: Optional[str] = None
-    classifications: List[Union[LBV1Text, LBV1Radio, LBV1Dropdown, LBV1Checklist]] = []
+    classifications: List[Union[LBV1Text, LBV1Radio, LBV1Dropdown,
+                                LBV1Checklist]] = []
+
     def dict(self, *args, **kwargs):
         res = super().dict(*args, **kwargs)
         # This means these are not video frames ..

labelbox/data/serialization/ndjson/classification.py

@@ -165,8 +165,8 @@ def to_common(
     def lookup_subclassification(
         annotation: ClassificationAnnotation
     ) -> Union[NDTextSubclass, NDChecklistSubclass, NDRadioSubclass]:
-        if isinstance(annotation, Dropdown):
-            raise TypeError("Dropdowns are not supported for MAL")
+        if isinstance(annotation.value, Dropdown):
+            raise TypeError("Dropdowns are not supported for MAL.")
         return {
             Text: NDTextSubclass,
             Checklist: NDChecklistSubclass,
@@ -213,13 +213,12 @@ def lookup_classification(
         annotation: Union[ClassificationAnnotation,
                           VideoClassificationAnnotation]
     ) -> Union[NDText, NDChecklist, NDRadio]:
-        if isinstance(annotation, Dropdown):
-            raise TypeError("Dropdowns are not supported for MAL")
+        if isinstance(annotation.value, Dropdown):
+            raise TypeError("Dropdowns are not supported for MAL.")
         return {
             Text: NDText,
             Checklist: NDChecklist,
-            Radio: NDRadio,
-            Dropdown: NDChecklist,
+            Radio: NDRadio
         }.get(type(annotation.value))
 
 

labelbox/data/serialization/ndjson/label.py

@@ -1,3 +1,4 @@
+import logging
 from itertools import groupby
 from labelbox.data.annotation_types.classification.classification import Dropdown
 from labelbox.data.annotation_types.metrics import ScalarMetric
@@ -17,6 +18,8 @@
 from .classification import NDChecklistSubclass, NDClassification, NDClassificationType, NDRadioSubclass
 from .objects import NDObject, NDObjectType
 
+logger = logging.getLogger(__name__)
+
 
 class NDLabel(BaseModel):
     annotations: List[Union[NDObjectType, NDClassificationType, NDMetricType]]
@@ -103,11 +106,6 @@ def _create_non_video_annotations(cls, label: Label):
         ]
         for annotation in non_video_annotations:
             if isinstance(annotation, ClassificationAnnotation):
-                if isinstance(annotation.value, Dropdown):
-                    raise ValueError(
-                        "Dropdowns are not supported by the NDJson format."
-                        " Please filter out Dropdown annotations before converting."
-                    )
                 yield NDClassification.from_common(annotation, label.data)
             elif isinstance(annotation, ObjectAnnotation):
                 yield NDObject.from_common(annotation, label.data)

tests/data/serialization/labelbox_v1/test_image.py

@@ -5,13 +5,10 @@
 from labelbox.data.serialization.labelbox_v1.converter import LBV1Converter
 
 
-
-@pytest.mark.parametrize(
-    "file_path",
-    [
-        'tests/data/assets/labelbox_v1/highly_nested_image.json',
-        'tests/data/assets/labelbox_v1/image_export.json'
-    ])
+@pytest.mark.parametrize("file_path", [
+    'tests/data/assets/labelbox_v1/highly_nested_image.json',
+    'tests/data/assets/labelbox_v1/image_export.json'
+])
 def test_image(file_path):
     with open(file_path, 'r') as file:
         payload = json.load(file)
@@ -30,10 +27,13 @@ def test_image(file_path):
                     # We don't add a classification key to the payload if there is no classifications.
                     annotation_a.pop('classifications')
 
-                if isinstance(annotation_b.get('classifications'), list) and len(annotation_b['classifications']):
+                if isinstance(annotation_b.get('classifications'),
+                              list) and len(annotation_b['classifications']):
                     if isinstance(annotation_b['classifications'][0], list):
-                        annotation_b['classifications'] = annotation_b['classifications'][0]
+                        annotation_b['classifications'] = annotation_b[
+                            'classifications'][0]
 
                 assert annotation_a == annotation_b
 
+
 # After check the nd serializer on this shit.. It should work for almost everything (except the other horse shit..)