wip

Author: Matt Sokoloff

labelbox/data/metrics/iou.py

@@ -5,7 +5,7 @@
 from itertools import product
 import numpy as np
 from collections import defaultdict
-
+from ..annotation_types.feature import FeatureSchema
 from ..annotation_types import (Label, ObjectAnnotation,
                                 ClassificationAnnotation, Mask, Geometry, Point,
                                 Line, Checklist, Text, Radio)
@@ -29,6 +29,7 @@ def subclass_ious(ground_truth: Label, prediction: Label) -> Dict[str, Optional[
     Otherwise this function is a bit naive and if you want something to specifically suite
     your use case then create a new function based off this one.
     """
+    identifying = get_identifying_key(prediction.annotations, ground_truth.annotations)
     prediction_annotations = _create_feature_lookup(prediction.annotations)
     ground_truth_annotations = _create_feature_lookup(ground_truth.annotations)
     feature_schemas = set(prediction_annotations.keys()).union(
@@ -48,15 +49,14 @@ def _create_classification_feature_lookup(annotations: Union[List[ObjectAnnotati
             classifications = [classification.answer.name or classification.answer.feature_schema_id for classification in classifications ]
             # TODO: create the lookup
             grouped_annotations[annotation.name or annotation.feature_schema_id].append(annotation)
-
         return grouped_annotations
 
 
-    ious = []
-    for key in feature_schemas:
+    #ious = []
+    #for key in feature_schemas:
         # We shouldn't have any nones. Since the keys are generated by the presence of the object.
-        prediction_annotations = prediction_annotations[key]
-        ground_truth_annotations =
+        #prediction_annotations = prediction_annotations[key]
+    #    #ground_truth_annotations =
 
 
 
@@ -314,34 +314,7 @@ def _create_feature_lookup(
 
     """
     # TODO: Add a check here.
-    """
-
-    We don't want to select name for one and then feature_schema_id for the other.
-    I think in another function we should check
-
-    Do we want to require that the user provides the feature name?
-    We don't really want schema ids showing up in the metric names..
-
-    So:
-
-    Also add a freakin test.
-    ####
-    all_schema_ids_defined_pred, all_names_defined_pred = check_references(pred_annotations)
-    if (not all_schema_ids_defined and not all_names_defined_pred):
-        raise ValueError("All data must have feature_schema_ids or names set")
-
 
-    all_schema_ids_defined_gt, all_names_defined_gt = check_references(gt_annotations)
-
-    #Prefer name becuse the user will be able to know what it means
-    #Schema id incase that doesn't exist..
-    if (all_names_defined_pred and all_names_defined_gt):
-        return 'name'
-    elif all_schema_ids_defined_pred and all_schema_ids_defined_gt:
-        return 'feature_schema_id'
-    else:
-        raise ValueError("Ground truth and prediction annotations must have set all name or feature ids. Otherwise there is no key to match on. Please update.")
-    """
     grouped_annotations = defaultdict(list)
     for annotation in annotations:
         grouped_annotations[annotation.name or
@@ -381,3 +354,46 @@ def _polygon_iou(poly1: Polygon, poly2: Polygon) -> float:
 def _mask_iou(mask1: np.ndarray, mask2: np.ndarray) -> float:
     """Computes iou between two binary segmentation masks."""
     return np.sum(mask1 & mask2) / np.sum(mask1 | mask2)
+
+
+def all_have_key(annotations: List[FeatureSchema]) -> Tuple[bool, bool]:
+    """
+    We want to make sure that all feature schemas have names set or feature_schema_ids set.
+
+    """
+    all_names = True
+    all_schemas = True
+    for annotation in annotations:
+        if annotation.name is None:
+            all_names = False
+        if annotation.feature_schema_id is None:
+            all_schemas = False
+    return all_schemas, all_names
+
+def get_identifying_key(pred_annotations, gt_annotations):
+    """
+    We don't want to select name for one and then feature_schema_id for the other.
+    I think in another function we should check
+
+    Do we want to require that the user provides the feature name?
+    We don't really want schema ids showing up in the metric names..
+
+    So:
+    """
+    #TODO: Also add a freakin test.
+    all_schema_ids_defined_pred, all_names_defined_pred = all_have_key(pred_annotations)
+    if (not all_schema_ids_defined_pred and not all_names_defined_pred):
+        raise ValueError("All data must have feature_schema_ids or names set")
+
+
+    all_schema_ids_defined_gt, all_names_defined_gt = all_have_key(gt_annotations)
+
+    #Prefer name becuse the user will be able to know what it means
+    #Schema id incase that doesn't exist..
+    if (all_names_defined_pred and all_names_defined_gt):
+        return 'name'
+    elif all_schema_ids_defined_pred and all_schema_ids_defined_gt:
+        return 'feature_schema_id'
+    else:
+        raise ValueError("Ground truth and prediction annotations must have set all name or feature ids. Otherwise there is no key to match on. Please update.")
+