feat(coco_utils): add scoring for classification model

src/nrtk_explorer/library/coco_utils.py

@@ -122,6 +122,18 @@ def keys_to_dataset_ids(image_dict):
     return {image_id_to_dataset_id(key): value for key, value in image_dict.items()}
 
 
+def make_cat_ids(dataset):
+    """Get category ids from annotations."""
+    label_to_id = {cat["name"]: cat["id"] for cat in dataset.cats.values()}
+
+    def get_cat_id(annotation):
+        if "category_id" in annotation:
+            return annotation["category_id"]
+        return label_to_id.get(annotation["label"], None)
+
+    return get_cat_id
+
+
 def compute_score(dataset, actual_info, predicted_info):
     """Compute score for image ids."""
 
@@ -153,6 +165,29 @@ def is_empty(prediction_pair):
 
     actual, predicted, ids = zip(*has_annotations)
 
+    all_annotations_have_bbox = all(
+        "bbox" in annotation
+        for annotation_list in actual + predicted
+        for annotation in annotation_list
+    )
+
+    if not all_annotations_have_bbox:
+        # score with classification method
+        get_cat_id = make_cat_ids(dataset)
+        for pair in has_annotations:
+            actual, predicted, id = pair
+            actual_cat_ids = [get_cat_id(annotation) for annotation in actual]
+            predicted_cat_ids = [get_cat_id(annotation) for annotation in predicted]
+            matching_cat_ids = sum(
+                1
+                for cat_id in predicted_cat_ids
+                if cat_id in actual_cat_ids and cat_id is not None
+            )
+            total_cat_ids = len(set(actual_cat_ids + predicted_cat_ids))
+            score = matching_cat_ids / total_cat_ids if total_cat_ids > 0 else 0.0
+            scores.append((id, score))
+        return scores
+
     if actual_info["type"] == "predictions":
         actual_converted = convert_from_predictions_to_first_arg(
             actual,