valor-lite 0.33.7__py3-none-any.whl → 0.33.8__py3-none-any.whl

valor_lite/detection/manager.py

@@ -1,9 +1,11 @@
+import warnings
 from collections import defaultdict
 from dataclasses import dataclass
+from typing import Type
 
 import numpy as np
+import valor_lite.detection.annotation as annotation
 from numpy.typing import NDArray
-from shapely.geometry import Polygon as ShapelyPolygon
 from tqdm import tqdm
 from valor_lite.detection.annotation import (
     Bitmask,
@@ -59,103 +61,6 @@ filtered_metrics = evaluator.evaluate(iou_thresholds=[0.5], filter_mask=filter_m
 """
 
 
-def _get_valor_dict_annotation_key(
-    annotation_type: type[BoundingBox] | type[Polygon] | type[Bitmask],
-) -> str:
-    """Get the correct JSON key to extract a given annotation type."""
-
-    if issubclass(annotation_type, BoundingBox):
-        return "bounding_box"
-    if issubclass(annotation_type, Polygon):
-        return "polygon"
-    else:
-        return "raster"
-
-
-def _get_annotation_representation(
-    annotation_type: type[BoundingBox] | type[Polygon] | type[Bitmask],
-) -> str:
-    """Get the correct representation of an annotation object."""
-
-    representation = (
-        "extrema"
-        if issubclass(annotation_type, BoundingBox)
-        else ("mask" if issubclass(annotation_type, Bitmask) else "shape")
-    )
-
-    return representation
-
-
-def _get_annotation_representation_from_valor_dict(
-    data: list,
-    annotation_type: type[BoundingBox] | type[Polygon] | type[Bitmask],
-) -> tuple[float, float, float, float] | ShapelyPolygon | NDArray[np.bool_]:
-    """Get the correct representation of an annotation object from a valor dictionary."""
-
-    if issubclass(annotation_type, BoundingBox):
-        x = [point[0] for shape in data for point in shape]
-        y = [point[1] for shape in data for point in shape]
-        return (min(x), max(x), min(y), max(y))
-    if issubclass(annotation_type, Polygon):
-        return ShapelyPolygon(data)
-    else:
-        return np.array(data)
-
-
-def _get_annotation_data(
-    keyed_groundtruths: dict,
-    keyed_predictions: dict,
-    annotation_type: type[BoundingBox] | type[Polygon] | type[Bitmask] | None,
-    key=int,
-) -> np.ndarray:
-    """Create an array of annotation pairs for use when calculating IOU. Needed because we unpack bounding box representations, but not bitmask or polygon representations."""
-    if annotation_type == BoundingBox:
-        return np.array(
-            [
-                np.array([*gextrema, *pextrema])
-                for _, _, _, pextrema in keyed_predictions[key]
-                for _, _, gextrema in keyed_groundtruths[key]
-            ]
-        )
-    else:
-        return np.array(
-            [
-                np.array([groundtruth_obj, prediction_obj])
-                for _, _, _, prediction_obj in keyed_predictions[key]
-                for _, _, groundtruth_obj in keyed_groundtruths[key]
-            ]
-        )
-
-
-def compute_iou(
-    data: NDArray[np.floating],
-    annotation_type: type[BoundingBox] | type[Polygon] | type[Bitmask],
-) -> NDArray[np.floating]:
-    """
-    Computes intersection-over-union (IoU) calculations for various annotation types.
-
-    Parameters
-    ----------
-    data : NDArray[np.floating]
-        A sorted array of bounding box, bitmask, or polygon pairs.
-    annotation_type: type[BoundingBox] | type[Polygon] | type[Bitmask]
-        The type of annotation contained in the data.
-
-
-    Returns
-    -------
-    NDArray[np.floating]
-        Computed IoU's.
-    """
-
-    if annotation_type == BoundingBox:
-        return compute_bbox_iou(data=data)
-    elif annotation_type == Bitmask:
-        return compute_bitmask_iou(data=data)
-    else:
-        return compute_polygon_iou(data=data)
-
-
 @dataclass
 class Filter:
     ranked_indices: NDArray[np.int32]
@@ -185,22 +90,17 @@ class Evaluator:
         self.prediction_examples: dict[int, NDArray[np.float16]] = dict()
 
         # label reference
-        self.label_to_index: dict[tuple[str, str], int] = dict()
-        self.index_to_label: dict[int, tuple[str, str]] = dict()
-
-        # label key reference
-        self.index_to_label_key: dict[int, str] = dict()
-        self.label_key_to_index: dict[str, int] = dict()
-        self.label_index_to_label_key_index: dict[int, int] = dict()
+        self.label_to_index: dict[str, int] = dict()
+        self.index_to_label: dict[int, str] = dict()
 
         # computation caches
-        self._detailed_pairs: NDArray[np.floating] = np.array([])
-        self._ranked_pairs: NDArray[np.floating] = np.array([])
+        self._detailed_pairs: NDArray[np.float64] = np.array([])
+        self._ranked_pairs: NDArray[np.float64] = np.array([])
         self._label_metadata: NDArray[np.int32] = np.array([])
         self._label_metadata_per_datum: NDArray[np.int32] = np.array([])
 
     @property
-    def ignored_prediction_labels(self) -> list[tuple[str, str]]:
+    def ignored_prediction_labels(self) -> list[str]:
         """
         Prediction labels that are not present in the ground truth set.
         """
@@ -211,7 +111,7 @@ class Evaluator:
         ]
 
     @property
-    def missing_prediction_labels(self) -> list[tuple[str, str]]:
+    def missing_prediction_labels(self) -> list[str]:
         """
         Ground truth labels that are not present in the prediction set.
         """
@@ -238,8 +138,7 @@ class Evaluator:
     def create_filter(
         self,
         datum_uids: list[str] | NDArray[np.int32] | None = None,
-        labels: list[tuple[str, str]] | NDArray[np.int32] | None = None,
-        label_keys: list[str] | NDArray[np.int32] | None = None,
+        labels: list[str] | NDArray[np.int32] | None = None,
     ) -> Filter:
         """
         Creates a filter that can be passed to an evaluation.
@@ -248,10 +147,8 @@ class Evaluator:
         ----------
         datum_uids : list[str] | NDArray[np.int32], optional
             An optional list of string uids or a numpy array of uid indices.
-        labels : list[tuple[str, str]] | NDArray[np.int32], optional
+        labels : list[str] | NDArray[np.int32], optional
             An optional list of labels or a numpy array of label indices.
-        label_keys : list[str] | NDArray[np.int32], optional
-            An optional list of label keys or a numpy array of label key indices.
 
         Returns
         -------
@@ -296,24 +193,6 @@ class Evaluator:
             ] = False
             mask_labels[~np.isin(np.arange(n_labels), labels)] = False
 
-        if label_keys is not None:
-            if isinstance(label_keys, list):
-                label_keys = np.array(
-                    [self.label_key_to_index[key] for key in label_keys]
-                )
-            label_indices = (
-                np.where(np.isclose(self._label_metadata[:, 2], label_keys))[0]
-                if label_keys.size > 0
-                else np.array([])
-            )
-            mask_ranked[
-                ~np.isin(self._ranked_pairs[:, 4].astype(int), label_indices)
-            ] = False
-            mask_detailed[
-                ~np.isin(self._detailed_pairs[:, 4].astype(int), label_indices)
-            ] = False
-            mask_labels[~np.isin(np.arange(n_labels), label_indices)] = False
-
         mask_label_metadata = (
             mask_datums[:, np.newaxis] & mask_labels[np.newaxis, :]
         )
@@ -321,13 +200,12 @@ class Evaluator:
         label_metadata_per_datum[:, ~mask_label_metadata] = 0
 
         label_metadata = np.zeros_like(self._label_metadata, dtype=np.int32)
-        label_metadata[:, :2] = np.transpose(
+        label_metadata = np.transpose(
             np.sum(
                 label_metadata_per_datum,
                 axis=1,
             )
         )
-        label_metadata[:, 2] = self._label_metadata[:, 2]
 
         return Filter(
             ranked_indices=np.where(mask_ranked)[0],
@@ -413,12 +291,10 @@ class Evaluator:
 
         metrics[MetricType.mAP] = [
             mAP(
-                value=mean_average_precision[iou_idx][label_key_idx],
+                value=mean_average_precision[iou_idx],
                 iou_threshold=iou_thresholds[iou_idx],
-                label_key=self.index_to_label_key[label_key_idx],
             )
             for iou_idx in range(mean_average_precision.shape[0])
-            for label_key_idx in range(mean_average_precision.shape[1])
         ]
 
         metrics[MetricType.APAveragedOverIOUs] = [
@@ -433,12 +309,8 @@ class Evaluator:
 
         metrics[MetricType.mAPAveragedOverIOUs] = [
             mAPAveragedOverIOUs(
-                value=mean_average_precision_average_over_ious[label_key_idx],
+                value=mean_average_precision_average_over_ious,
                 iou_thresholds=iou_thresholds,
-                label_key=self.index_to_label_key[label_key_idx],
-            )
-            for label_key_idx in range(
-                mean_average_precision_average_over_ious.shape[0]
             )
         ]
 
@@ -456,13 +328,11 @@ class Evaluator:
 
         metrics[MetricType.mAR] = [
             mAR(
-                value=mean_average_recall[score_idx][label_key_idx],
+                value=mean_average_recall[score_idx],
                 iou_thresholds=iou_thresholds,
                 score_threshold=score_thresholds[score_idx],
-                label_key=self.index_to_label_key[label_key_idx],
             )
             for score_idx in range(mean_average_recall.shape[0])
-            for label_key_idx in range(mean_average_recall.shape[1])
         ]
 
         metrics[MetricType.ARAveragedOverScores] = [
@@ -478,13 +348,9 @@ class Evaluator:
 
         metrics[MetricType.mARAveragedOverScores] = [
             mARAveragedOverScores(
-                value=mean_average_recall_averaged_over_scores[label_key_idx],
+                value=mean_average_recall_averaged_over_scores,
                 score_thresholds=score_thresholds,
                 iou_thresholds=iou_thresholds,
-                label_key=self.index_to_label_key[label_key_idx],
-            )
-            for label_key_idx in range(
-                mean_average_recall_averaged_over_scores.shape[0]
             )
         ]
 
@@ -570,10 +436,22 @@ class Evaluator:
 
         return metrics
 
+    def _convert_example_to_dict(
+        self, box: NDArray[np.float16]
+    ) -> dict[str, float]:
+        """
+        Converts a cached bounding box example to dictionary format.
+        """
+        return {
+            "xmin": box[0],
+            "xmax": box[1],
+            "ymin": box[2],
+            "ymax": box[3],
+        }
+
     def _unpack_confusion_matrix(
         self,
-        confusion_matrix: NDArray[np.floating],
-        label_key_idx: int,
+        confusion_matrix: NDArray[np.float64],
         number_of_labels: int,
         number_of_examples: int,
     ) -> dict[
@@ -586,7 +464,7 @@ class Evaluator:
                 | list[
                     dict[
                         str,
-                        str | float | tuple[float, float, float, float],
+                        str | float | dict[str, float],
                     ]
                 ],
             ],
@@ -629,8 +507,8 @@ class Evaluator:
         )
 
         return {
-            self.index_to_label[gt_label_idx][1]: {
-                self.index_to_label[pd_label_idx][1]: {
+            self.index_to_label[gt_label_idx]: {
+                self.index_to_label[pd_label_idx]: {
                     "count": max(
                         int(confusion_matrix[gt_label_idx, pd_label_idx, 0]),
                         0,
@@ -642,7 +520,7 @@ class Evaluator:
                                     gt_label_idx, pd_label_idx, example_idx
                                 )
                             ],
-                            "groundtruth": tuple(
+                            "groundtruth": self._convert_example_to_dict(
                                 self.groundtruth_examples[
                                     datum_idx(
                                         gt_label_idx,
@@ -655,9 +533,9 @@ class Evaluator:
                                         pd_label_idx,
                                         example_idx,
                                     )
-                                ].tolist()
+                                ]
                             ),
-                            "prediction": tuple(
+                            "prediction": self._convert_example_to_dict(
                                 self.prediction_examples[
                                     datum_idx(
                                         gt_label_idx,
@@ -670,7 +548,7 @@ class Evaluator:
                                         pd_label_idx,
                                         example_idx,
                                     )
-                                ].tolist()
+                                ]
                             ),
                             "score": score_idx(
                                 gt_label_idx, pd_label_idx, example_idx
@@ -682,30 +560,20 @@ class Evaluator:
                     ],
                 }
                 for pd_label_idx in range(number_of_labels)
-                if (
-                    self.label_index_to_label_key_index[pd_label_idx]
-                    == label_key_idx
-                )
             }
             for gt_label_idx in range(number_of_labels)
-            if (
-                self.label_index_to_label_key_index[gt_label_idx]
-                == label_key_idx
-            )
         }
 
     def _unpack_hallucinations(
         self,
-        hallucinations: NDArray[np.floating],
-        label_key_idx: int,
+        hallucinations: NDArray[np.float64],
         number_of_labels: int,
         number_of_examples: int,
     ) -> dict[
         str,
         dict[
             str,
-            int
-            | list[dict[str, str | float | tuple[float, float, float, float]]],
+            int | list[dict[str, str | float | dict[str, float]]],
         ],
     ]:
         """
@@ -740,7 +608,7 @@ class Evaluator:
         )
 
         return {
-            self.index_to_label[pd_label_idx][1]: {
+            self.index_to_label[pd_label_idx]: {
                 "count": max(
                     int(hallucinations[pd_label_idx, 0]),
                     0,
@@ -750,12 +618,10 @@ class Evaluator:
                         "datum": self.index_to_uid[
                             datum_idx(pd_label_idx, example_idx)
                         ],
-                        "prediction": tuple(
+                        "prediction": self._convert_example_to_dict(
                             self.prediction_examples[
                                 datum_idx(pd_label_idx, example_idx)
-                            ][
-                                prediction_idx(pd_label_idx, example_idx)
-                            ].tolist()
+                            ][prediction_idx(pd_label_idx, example_idx)]
                         ),
                         "score": score_idx(pd_label_idx, example_idx),
                     }
@@ -764,25 +630,14 @@ class Evaluator:
                 ],
             }
             for pd_label_idx in range(number_of_labels)
-            if (
-                self.label_index_to_label_key_index[pd_label_idx]
-                == label_key_idx
-            )
         }
 
     def _unpack_missing_predictions(
         self,
         missing_predictions: NDArray[np.int32],
-        label_key_idx: int,
         number_of_labels: int,
         number_of_examples: int,
-    ) -> dict[
-        str,
-        dict[
-            str,
-            int | list[dict[str, str | tuple[float, float, float, float]]],
-        ],
-    ]:
+    ) -> dict[str, dict[str, int | list[dict[str, str | dict[str, float]]]]]:
         """
         Unpacks a numpy array of missing prediction counts and examples.
         """
@@ -806,7 +661,7 @@ class Evaluator:
         )
 
         return {
-            self.index_to_label[gt_label_idx][1]: {
+            self.index_to_label[gt_label_idx]: {
                 "count": max(
                     int(missing_predictions[gt_label_idx, 0]),
                     0,
@@ -816,12 +671,10 @@ class Evaluator:
                         "datum": self.index_to_uid[
                             datum_idx(gt_label_idx, example_idx)
                         ],
-                        "groundtruth": tuple(
+                        "groundtruth": self._convert_example_to_dict(
                             self.groundtruth_examples[
                                 datum_idx(gt_label_idx, example_idx)
-                            ][
-                                groundtruth_idx(gt_label_idx, example_idx)
-                            ].tolist()
+                            ][groundtruth_idx(gt_label_idx, example_idx)]
                         ),
                     }
                     for example_idx in range(number_of_examples)
@@ -829,15 +682,11 @@ class Evaluator:
                 ],
             }
             for gt_label_idx in range(number_of_labels)
-            if (
-                self.label_index_to_label_key_index[gt_label_idx]
-                == label_key_idx
-            )
         }
 
     def _compute_confusion_matrix(
         self,
-        data: NDArray[np.floating],
+        data: NDArray[np.float64],
         label_metadata: NDArray[np.int32],
         iou_thresholds: list[float],
         score_thresholds: list[float],
@@ -848,7 +697,7 @@ class Evaluator:
 
         Parameters
         ----------
-        data : NDArray[np.floating]
+        data : NDArray[np.float64]
             An array containing detailed pairs of detections.
         label_metadata : NDArray[np.int32]
             An array containing label metadata.
@@ -885,19 +734,16 @@ class Evaluator:
             ConfusionMatrix(
                 iou_threshold=iou_thresholds[iou_idx],
                 score_threshold=score_thresholds[score_idx],
-                label_key=label_key,
                 number_of_examples=number_of_examples,
                 confusion_matrix=self._unpack_confusion_matrix(
                     confusion_matrix=confusion_matrix[
                         iou_idx, score_idx, :, :, :
                     ],
-                    label_key_idx=label_key_idx,
                     number_of_labels=n_labels,
                     number_of_examples=number_of_examples,
                 ),
                 hallucinations=self._unpack_hallucinations(
                     hallucinations=hallucinations[iou_idx, score_idx, :, :],
-                    label_key_idx=label_key_idx,
                     number_of_labels=n_labels,
                     number_of_examples=number_of_examples,
                 ),
@@ -905,12 +751,10 @@ class Evaluator:
                     missing_predictions=missing_predictions[
                         iou_idx, score_idx, :, :
                     ],
-                    label_key_idx=label_key_idx,
                     number_of_labels=n_labels,
                     number_of_examples=number_of_examples,
                 ),
             )
-            for label_key_idx, label_key in self.index_to_label_key.items()
             for iou_idx in range(n_ious)
             for score_idx in range(n_scores)
         ]
@@ -923,7 +767,7 @@ class DataLoader:
 
     def __init__(self):
         self._evaluator = Evaluator()
-        self.pairs: list[NDArray[np.floating]] = list()
+        self.pairs: list[NDArray[np.float64]] = list()
         self.groundtruth_count = defaultdict(lambda: defaultdict(int))
         self.prediction_count = defaultdict(lambda: defaultdict(int))
 
@@ -947,51 +791,36 @@ class DataLoader:
             self._evaluator.index_to_uid[index] = uid
         return self._evaluator.uid_to_index[uid]
 
-    def _add_label(self, label: tuple[str, str]) -> tuple[int, int]:
+    def _add_label(self, label: str) -> int:
         """
         Helper function for adding a label to the cache.
 
         Parameters
         ----------
-        label : tuple[str, str]
-            The label as a tuple in format (key, value).
+        label : str
+            The label associated with the annotation.
 
         Returns
         -------
         int
             Label index.
-        int
-            Label key index.
         """
 
         label_id = len(self._evaluator.index_to_label)
-        label_key_id = len(self._evaluator.index_to_label_key)
         if label not in self._evaluator.label_to_index:
             self._evaluator.label_to_index[label] = label_id
             self._evaluator.index_to_label[label_id] = label
 
-            # update label key index
-            if label[0] not in self._evaluator.label_key_to_index:
-                self._evaluator.label_key_to_index[label[0]] = label_key_id
-                self._evaluator.index_to_label_key[label_key_id] = label[0]
-                label_key_id += 1
-
-            self._evaluator.label_index_to_label_key_index[
-                label_id
-            ] = self._evaluator.label_key_to_index[label[0]]
             label_id += 1
 
-        return (
-            self._evaluator.label_to_index[label],
-            self._evaluator.label_key_to_index[label[0]],
-        )
+        return self._evaluator.label_to_index[label]
 
     def _compute_ious_and_cache_pairs(
         self,
         uid_index: int,
-        keyed_groundtruths: dict,
-        keyed_predictions: dict,
-        annotation_type: type[BoundingBox] | type[Polygon] | type[Bitmask],
+        groundtruths: list,
+        predictions: list,
+        annotation_type: Type[BoundingBox] | Type[Polygon] | Type[Bitmask],
     ) -> None:
         """
         Compute IOUs between groundtruths and preditions before storing as pairs.
@@ -1000,122 +829,92 @@ class DataLoader:
         ----------
         uid_index: int
             The index of the detection.
-        keyed_groundtruths: dict
-            A dictionary of groundtruths.
-        keyed_predictions: dict
-            A dictionary of predictions.
+        groundtruths: list
+            A list of groundtruths.
+        predictions: list
+            A list of predictions.
         annotation_type: type[BoundingBox] | type[Polygon] | type[Bitmask]
             The type of annotation to compute IOUs for.
         """
-        gt_keys = set(keyed_groundtruths.keys())
-        pd_keys = set(keyed_predictions.keys())
-        joint_keys = gt_keys.intersection(pd_keys)
-        gt_unique_keys = gt_keys - pd_keys
-        pd_unique_keys = pd_keys - gt_keys
 
         pairs = list()
-        for key in joint_keys:
-            n_predictions = len(keyed_predictions[key])
-            n_groundtruths = len(keyed_groundtruths[key])
-            data = _get_annotation_data(
-                keyed_groundtruths=keyed_groundtruths,
-                keyed_predictions=keyed_predictions,
-                key=key,
-                annotation_type=annotation_type,
-            )
-            ious = compute_iou(data=data, annotation_type=annotation_type)
-            mask_nonzero_iou = (ious > 1e-9).reshape(
-                (n_predictions, n_groundtruths)
-            )
-            mask_ious_halluc = ~(mask_nonzero_iou.any(axis=1))
-            mask_ious_misprd = ~(mask_nonzero_iou.any(axis=0))
+        n_predictions = len(predictions)
+        n_groundtruths = len(groundtruths)
 
-            pairs.extend(
-                [
-                    np.array(
-                        [
-                            float(uid_index),
-                            float(gidx),
-                            float(pidx),
-                            ious[pidx * len(keyed_groundtruths[key]) + gidx],
-                            float(glabel),
-                            float(plabel),
-                            float(score),
-                        ]
-                    )
-                    for pidx, plabel, score, _ in keyed_predictions[key]
-                    for gidx, glabel, _ in keyed_groundtruths[key]
-                    if ious[pidx * len(keyed_groundtruths[key]) + gidx] > 1e-9
-                ]
-            )
-            pairs.extend(
-                [
-                    np.array(
-                        [
-                            float(uid_index),
-                            -1.0,
-                            float(pidx),
-                            0.0,
-                            -1.0,
-                            float(plabel),
-                            float(score),
-                        ]
-                    )
-                    for pidx, plabel, score, _ in keyed_predictions[key]
-                    if mask_ious_halluc[pidx]
-                ]
-            )
-            pairs.extend(
-                [
-                    np.array(
-                        [
-                            float(uid_index),
-                            float(gidx),
-                            -1.0,
-                            0.0,
-                            float(glabel),
-                            -1.0,
-                            -1.0,
-                        ]
-                    )
-                    for gidx, glabel, _ in keyed_groundtruths[key]
-                    if mask_ious_misprd[gidx]
-                ]
-            )
-        for key in gt_unique_keys:
-            pairs.extend(
-                [
-                    np.array(
-                        [
-                            float(uid_index),
-                            float(gidx),
-                            -1.0,
-                            0.0,
-                            float(glabel),
-                            -1.0,
-                            -1.0,
-                        ]
-                    )
-                    for gidx, glabel, _ in keyed_groundtruths[key]
-                ]
-            )
-        for key in pd_unique_keys:
-            pairs.extend(
-                [
-                    np.array(
-                        [
-                            float(uid_index),
-                            -1.0,
-                            float(pidx),
-                            0.0,
-                            -1.0,
-                            float(plabel),
-                            float(score),
-                        ]
-                    )
-                    for pidx, plabel, score, _ in keyed_predictions[key]
-                ]
-            )
+        all_pairs = np.array(
+            [
+                np.array([gann, pann])
+                for _, _, _, pann in predictions
+                for _, _, gann in groundtruths
+            ]
+        )
+
+        match annotation_type:
+            case annotation.BoundingBox:
+                ious = compute_bbox_iou(all_pairs)
+            case annotation.Polygon:
+                ious = compute_polygon_iou(all_pairs)
+            case annotation.Bitmask:
+                ious = compute_bitmask_iou(all_pairs)
+            case _:
+                raise ValueError(
+                    f"Invalid annotation type `{annotation_type}`."
+                )
+
+        ious = ious.reshape(n_predictions, n_groundtruths)
+        predictions_with_iou_of_zero = np.where((ious < 1e-9).all(axis=1))[0]
+        groundtruths_with_iou_of_zero = np.where((ious < 1e-9).all(axis=0))[0]
+
+        pairs.extend(
+            [
+                np.array(
+                    [
+                        float(uid_index),
+                        float(gidx),
+                        float(pidx),
+                        ious[pidx, gidx],
+                        float(glabel),
+                        float(plabel),
+                        float(score),
+                    ]
+                )
+                for pidx, plabel, score, _ in predictions
+                for gidx, glabel, _ in groundtruths
+                if ious[pidx, gidx] >= 1e-9
+            ]
+        )
+        pairs.extend(
+            [
+                np.array(
+                    [
+                        float(uid_index),
+                        -1.0,
+                        float(predictions[index][0]),
+                        0.0,
+                        -1.0,
+                        float(predictions[index][1]),
+                        float(predictions[index][2]),
+                    ]
+                )
+                for index in predictions_with_iou_of_zero
+            ]
+        )
+        pairs.extend(
+            [
+                np.array(
+                    [
+                        float(uid_index),
+                        float(groundtruths[index][0]),
+                        -1.0,
+                        0.0,
+                        float(groundtruths[index][1]),
+                        -1.0,
+                        -1.0,
+                    ]
+                )
+                for index in groundtruths_with_iou_of_zero
+            ]
+        )
 
         self.pairs.append(np.array(pairs))
 
@@ -1157,12 +956,8 @@ class DataLoader:
             )
 
             # cache labels and annotations
-            keyed_groundtruths = defaultdict(list)
-            keyed_predictions = defaultdict(list)
-
-            representation_property = _get_annotation_representation(
-                annotation_type=annotation_type
-            )
+            groundtruths = list()
+            predictions = list()
 
             for gidx, gann in enumerate(detection.groundtruths):
                 if not isinstance(gann, annotation_type):
@@ -1170,26 +965,17 @@ class DataLoader:
                         f"Expected {annotation_type}, but annotation is of type {type(gann)}."
                     )
 
-                if isinstance(gann, BoundingBox):
-                    self._evaluator.groundtruth_examples[uid_index][
-                        gidx
-                    ] = getattr(gann, representation_property)
-                else:
-                    converted_box = gann.to_box()
-                    self._evaluator.groundtruth_examples[uid_index][gidx] = (
-                        getattr(converted_box, "extrema")
-                        if converted_box is not None
-                        else None
-                    )
+                self._evaluator.groundtruth_examples[uid_index][
+                    gidx
+                ] = gann.extrema
                 for glabel in gann.labels:
-                    label_idx, label_key_idx = self._add_label(glabel)
+                    label_idx = self._add_label(glabel)
                     self.groundtruth_count[label_idx][uid_index] += 1
-                    representation = getattr(gann, representation_property)
-                    keyed_groundtruths[label_key_idx].append(
+                    groundtruths.append(
                         (
                             gidx,
                             label_idx,
-                            representation,
+                            gann.annotation,
                         )
                     )
 
@@ -1199,36 +985,25 @@ class DataLoader:
                         f"Expected {annotation_type}, but annotation is of type {type(pann)}."
                     )
 
-                if isinstance(pann, BoundingBox):
-                    self._evaluator.prediction_examples[uid_index][
-                        pidx
-                    ] = getattr(pann, representation_property)
-                else:
-                    converted_box = pann.to_box()
-                    self._evaluator.prediction_examples[uid_index][pidx] = (
-                        getattr(converted_box, "extrema")
-                        if converted_box is not None
-                        else None
-                    )
+                self._evaluator.prediction_examples[uid_index][
+                    pidx
+                ] = pann.extrema
                 for plabel, pscore in zip(pann.labels, pann.scores):
-                    label_idx, label_key_idx = self._add_label(plabel)
+                    label_idx = self._add_label(plabel)
                     self.prediction_count[label_idx][uid_index] += 1
-                    representation = representation = getattr(
-                        pann, representation_property
-                    )
-                    keyed_predictions[label_key_idx].append(
+                    predictions.append(
                         (
                             pidx,
                             label_idx,
                             pscore,
-                            representation,
+                            pann.annotation,
                         )
                     )
 
             self._compute_ious_and_cache_pairs(
                 uid_index=uid_index,
-                keyed_groundtruths=keyed_groundtruths,
-                keyed_predictions=keyed_predictions,
+                groundtruths=groundtruths,
+                predictions=predictions,
                 annotation_type=annotation_type,
             )
 
@@ -1295,10 +1070,9 @@ class DataLoader:
             annotation_type=Bitmask,
         )
 
-    def _add_data_from_valor_dict(
+    def add_bounding_boxes_from_valor_dict(
         self,
         detections: list[tuple[dict, dict]],
-        annotation_type: type[Bitmask] | type[BoundingBox] | type[Polygon],
         show_progress: bool = False,
     ):
         """
@@ -1313,9 +1087,29 @@ class DataLoader:
         show_progress : bool, default=False
             Toggle for tqdm progress bar.
         """
+        warnings.warn(
+            "The `...from_valor_dict` functions are deprecated and will be deleted in the near future. Use `add_bounding_boxes`, `add_bitmasks`, or `add_polygons` instead.",
+            DeprecationWarning,
+        )
+
+        def _get_bbox_extrema(
+            data: list,
+        ) -> tuple[float, float, float, float]:
+            """Get the correct representation of an annotation object from a valor dictionary."""
+            x = [point[0] for shape in data for point in shape]
+            y = [point[1] for shape in data for point in shape]
+            return (min(x), max(x), min(y), max(y))
 
         disable_tqdm = not show_progress
         for groundtruth, prediction in tqdm(detections, disable=disable_tqdm):
+
+            if not isinstance(groundtruth, dict) or not isinstance(
+                prediction, dict
+            ):
+                raise ValueError(
+                    f"Received values with type `{type(groundtruth)}` which are not valid Valor dictionaries."
+                )
+
             # update metadata
             self._evaluator.n_datums += 1
             self._evaluator.n_groundtruths += len(groundtruth["annotations"])
@@ -1333,112 +1127,63 @@ class DataLoader:
             )
 
             # cache labels and annotations
-            keyed_groundtruths = defaultdict(list)
-            keyed_predictions = defaultdict(list)
-
-            annotation_key = _get_valor_dict_annotation_key(
-                annotation_type=annotation_type
-            )
-            invalid_keys = list(
-                filter(
-                    lambda x: x != annotation_key,
-                    ["bounding_box", "raster", "polygon"],
-                )
-            )
+            groundtruths = list()
+            predictions = list()
 
             for gidx, gann in enumerate(groundtruth["annotations"]):
-                if (gann[annotation_key] is None) or any(
-                    [gann[k] is not None for k in invalid_keys]
-                ):
+                if gann["bounding_box"] is None:
                     raise ValueError(
-                        f"Input JSON doesn't contain {annotation_type} data, or contains data for multiple annotation types."
-                    )
-                if annotation_type == BoundingBox:
-                    self._evaluator.groundtruth_examples[uid_index][
-                        gidx
-                    ] = np.array(
-                        _get_annotation_representation_from_valor_dict(
-                            gann[annotation_key],
-                            annotation_type=annotation_type,
-                        ),
+                        f"Detection `{groundtruth['datum']['uid']}` contains a ground truth without a bounding box."
                     )
-
+                self._evaluator.groundtruth_examples[uid_index][
+                    gidx
+                ] = np.array(
+                    _get_bbox_extrema(gann["bounding_box"]),
+                )
                 for valor_label in gann["labels"]:
-                    glabel = (valor_label["key"], valor_label["value"])
-                    label_idx, label_key_idx = self._add_label(glabel)
+                    if valor_label["key"] != "name":
+                        continue
+                    glabel = f'{valor_label["key" ]}_{valor_label[ "value" ]}'
+                    label_idx = self._add_label(glabel)
                     self.groundtruth_count[label_idx][uid_index] += 1
-                    keyed_groundtruths[label_key_idx].append(
+                    groundtruths.append(
                         (
                             gidx,
                             label_idx,
-                            _get_annotation_representation_from_valor_dict(
-                                gann[annotation_key],
-                                annotation_type=annotation_type,
-                            ),
+                            _get_bbox_extrema(gann["bounding_box"]),
                         )
                     )
             for pidx, pann in enumerate(prediction["annotations"]):
-                if (pann[annotation_key] is None) or any(
-                    [pann[k] is not None for k in invalid_keys]
-                ):
+                if pann["bounding_box"] is None:
                     raise ValueError(
-                        f"Input JSON doesn't contain {annotation_type} data, or contains data for multiple annotation types."
-                    )
-
-                if annotation_type == BoundingBox:
-                    self._evaluator.prediction_examples[uid_index][
-                        pidx
-                    ] = np.array(
-                        _get_annotation_representation_from_valor_dict(
-                            pann[annotation_key],
-                            annotation_type=annotation_type,
-                        )
+                        f"Detection `{prediction['datum']['uid']}` contains a prediction without a bounding box."
                     )
+                self._evaluator.prediction_examples[uid_index][
+                    pidx
+                ] = np.array(_get_bbox_extrema(pann["bounding_box"]))
                 for valor_label in pann["labels"]:
-                    plabel = (valor_label["key"], valor_label["value"])
+                    if valor_label["key"] != "name":
+                        continue
+                    plabel = valor_label["value"]
                     pscore = valor_label["score"]
-                    label_idx, label_key_idx = self._add_label(plabel)
+                    label_idx = self._add_label(plabel)
                     self.prediction_count[label_idx][uid_index] += 1
-                    keyed_predictions[label_key_idx].append(
+                    predictions.append(
                         (
                             pidx,
                             label_idx,
                             pscore,
-                            _get_annotation_representation_from_valor_dict(
-                                pann[annotation_key],
-                                annotation_type=annotation_type,
-                            ),
+                            _get_bbox_extrema(pann["bounding_box"]),
                         )
                     )
 
             self._compute_ious_and_cache_pairs(
                 uid_index=uid_index,
-                keyed_groundtruths=keyed_groundtruths,
-                keyed_predictions=keyed_predictions,
-                annotation_type=annotation_type,
+                groundtruths=groundtruths,
+                predictions=predictions,
+                annotation_type=BoundingBox,
             )
 
-    def add_bounding_boxes_from_valor_dict(
-        self,
-        detections: list[tuple[dict, dict]],
-        show_progress: bool = False,
-    ):
-        """
-        Adds Valor-format bounding box detections to the cache.
-
-        Parameters
-        ----------
-        detections : list[tuple[dict, dict]]
-            A list of groundtruth, prediction pairs in Valor-format dictionaries.
-        show_progress : bool, default=False
-            Toggle for tqdm progress bar.
-        """
-        return self._add_data_from_valor_dict(
-            detections=detections,
-            show_progress=show_progress,
-            annotation_type=BoundingBox,
-        )
-
     def finalize(self) -> Evaluator:
         """
         Performs data finalization and some preprocessing steps.
@@ -1494,11 +1239,6 @@ class DataLoader:
                             ]
                         )
                     ),
-                    float(
-                        self._evaluator.label_index_to_label_key_index[
-                            label_idx
-                        ]
-                    ),
                 ]
                 for label_idx in range(n_labels)
             ]