Metrics docs overhaul

Author: LinasKo

docs/detection/metrics.md renamed to docs/detection/legacy_metrics.md

@@ -2,7 +2,10 @@
 comments: true
 ---
 
-# Metrics
+# Legacy Metrics
+
+Starting with `0.23.0`, a new metrics module is being introduced to supervision.
+Metrics here are part of the legacy evaluation API and will be deprecated in the future.
 
 <div class="md-typeset">
     <h2><a href="#supervision.metrics.detection.ConfusionMatrix">ConfusionMatrix</a></h2>

docs/metrics/intersection_over_union.md

@@ -3,6 +3,16 @@ comments: true
 status: new
 ---
 
-# Detections
+# Intersection over Union
+
+<div class="md-typeset">
+    <h2><a href="#supervision.metrics.intersection_over_union.IntersectionOverUnion">IntersectionOverUnion</a></h2>
+</div>
 
 :::supervision.metrics.intersection_over_union.IntersectionOverUnion
+
+<div class="md-typeset">
+    <h2><a href="#supervision.metrics.intersection_over_union.IntersectionOverUnionResult">IntersectionOverUnionResult</a></h2>
+</div>
+
+:::supervision.metrics.intersection_over_union.IntersectionOverUnionResult

docs/metrics/mean_average_precision.md

@@ -0,0 +1,18 @@
+---
+comments: true
+status: new
+---
+
+# Mean Average Precision
+
+<div class="md-typeset">
+    <h2><a href="#supervision.metrics.mean_average_precision.MeanAveragePrecision">MeanAveragePrecision</a></h2>
+</div>
+
+:::supervision.metrics.mean_average_precision.MeanAveragePrecision
+
+<div class="md-typeset">
+    <h2><a href="#supervision.metrics.mean_average_precision.MeanAveragePrecisionResult">MeanAveragePrecisionResult</a></h2>
+</div>
+
+:::supervision.metrics.mean_average_precision.MeanAveragePrecisionResult

mkdocs.yml

@@ -47,7 +47,6 @@ nav:
       - Detection and Segmentation:
           - Core: detection/core.md
           - Annotators: detection/annotators.md
-          - Metrics: detection/metrics.md
           - Double Detection Filter: detection/double_detection_filter.md
           - Utils: detection/utils.md
       - Keypoint Detection:
@@ -67,6 +66,8 @@ nav:
           - Utils: datasets/utils.md
       - Metrics:
           - IoU: metrics/intersection_over_union.md
+          - mAP: metrics/mean_average_precision.md
+          - Legacy Metrics: detection/legacy_metrics.md
       - Utils:
           - Video: utils/video.md
           - Image: utils/image.md

supervision/metrics/intersection_over_union.py

@@ -65,15 +65,13 @@ def update(
     ) -> IntersectionOverUnion:
         """
         Add data to the metric, without computing the result.
-        Should call all update methods of the shared data store.
 
         Args:
             data_1 (Union[Detection, List[Detections]]): The first set of data.
             data_2 (Union[Detection, List[Detections]]): The second set of data.
 
         Returns:
-            Metric: The metric object itself. You can get the metric result
-            by calling the `compute` method.
+            (IntersectionOverUnion): The updated metric instance.
         """
         if self._is_store_shared:
             # Should be updated by the parent metric
@@ -103,10 +101,32 @@ def compute(self) -> IntersectionOverUnionResult:
         Uses the data set with the `update` method.
 
         Returns:
-            Dict[int, npt.NDArray[np.float32]]: A dictionary with class IDs as keys.
-            If no class ID is provided, the key is the value CLASS_ID_NONE. The values
-            are (N, M) arrays where N is the number of predictions and M is the number
-            of targets.
+            IntersectionOverUnionResult: IoU results.
+
+        Example:
+            ```python
+            import supervision as sv
+            from supervision.metrics import IntersectionOverUnion
+
+            detections_1 = sv.Detections(...)
+            detections_2 = sv.Detections(...)
+
+            iou_metric = IntersectionOverUnion(class_agnostic=False)
+            iou_result = map_metric.update(detections_1, detections_2).compute()
+            print(iou_result)
+
+            class_id = 2
+            ious = iou_result[class_id]
+
+            class_id = -1  # no class
+            ious = iou_result[class_id]
+
+            for class_id, ious in iou_result:
+                ...
+
+            iou_result.plot()
+            ```
+
         """
         ious_by_class = {}
         for class_id in self._store.get_classes():
@@ -135,25 +155,75 @@ def compute(self) -> IntersectionOverUnionResult:
 @dataclass
 class IntersectionOverUnionResult:
     ious_by_class: Dict[int, npt.NDArray[np.float32]]
+    """The IoU matrices for each class."""
+
     metric_target: MetricTarget
+    """
+    Defines the type of data used for the metric - boxes, masks or
+    oriented bounding boxes.
+    """
 
     @property
     def class_ids(self) -> List[int]:
         return list(self.ious_by_class.keys())
 
     def __getitem__(self, class_id: int) -> npt.NDArray[np.float32]:
+        """
+        Get the IoU matrix for a specific class.
+
+        Args:
+            class_id (int): The class ID. Set `-1` to access "no class" data.
+            If class-agnostic IoU was used, all class IDs will be `-1`.
+
+        Returns:
+            (npt.NDArray[np.float32]): The IoU matrix for the class.
+
+        Example:
+            ```python
+            class_id = 2
+            ious = iou_result[class_id]
+            ```
+        """
         return self.ious_by_class[class_id]
 
     def __iter__(self):
+        """
+        Iterate over the IoU matrices for each class.
+
+        Returns:
+            (Iterator[Tuple[int, npt.NDArray[np.float32]]]): An iterator
+                with class IDs as keys and IoU matrices as values.
+
+        Example:
+            ```python
+            for class_id, ious in iou_result:
+                ...
+            ```
+        """
         return iter(self.ious_by_class.items())
 
     def __str__(self) -> str:
+        """
+        Format the IoU results as a pretty string.
+
+        Example:
+            ```python
+            print(iou_result)
+            ```
+        """
         out_str = f"{self.__class__.__name__}:\n"
         for class_id, iou in self.ious_by_class.items():
             out_str += f"IoUs for class {class_id}:\n{str(iou)}\n"
         return out_str
 
     def to_pandas(self) -> Dict[int, "pd.DataFrame"]:
+        """
+        Convert the results to multiple pandas DataFrames.
+
+        Returns:
+            (Dict[int, pd.DataFrame]): A dictionary with class IDs as keys and pandas
+                DataFrames as values.
+        """
         ensure_pandas_installed()
         import pandas as pd
 

supervision/metrics/mean_average_precision.py

@@ -52,6 +52,16 @@ def update(
         predictions: Union[Detections, List[Detections]],
         targets: Union[Detections, List[Detections]],
     ) -> MeanAveragePrecision:
+        """
+        Add new predictions and targets to the metric, but do not compute the result.
+
+        Args:
+            predictions (Union[Detections, List[Detections]]): The predicted detections.
+            targets (Union[Detections, List[Detections]]): The ground-truth detections.
+
+        Returns:
+            (MeanAveragePrecision): The updated metric instance.
+        """
         if not isinstance(predictions, list):
             predictions = [predictions]
         if not isinstance(targets, list):
@@ -88,45 +98,22 @@ def compute(
                 number of ground-truth objects. Each row is expected to be in
                 `(x_min, y_min, x_max, y_max, class)` format.
         Returns:
-            MeanAveragePrecision: New instance of MeanAveragePrecision.
+            (MeanAveragePrecision): New instance of MeanAveragePrecision.
 
         Example:
             ```python
             import supervision as sv
-            import numpy as np
-
-            targets = (
-                [
-                    np.array(
-                        [
-                            [0.0, 0.0, 3.0, 3.0, 1],
-                            [2.0, 2.0, 5.0, 5.0, 1],
-                            [6.0, 1.0, 8.0, 3.0, 2],
-                        ]
-                    ),
-                    np.array([[1.0, 1.0, 2.0, 2.0, 2]]),
-                ]
-            )
+            from supervision.metrics import MeanAveragePrecision
 
-            predictions = [
-                np.array(
-                    [
-                        [0.0, 0.0, 3.0, 3.0, 1, 0.9],
-                        [0.1, 0.1, 3.0, 3.0, 0, 0.9],
-                        [6.0, 1.0, 8.0, 3.0, 1, 0.8],
-                        [1.0, 6.0, 2.0, 7.0, 1, 0.8],
-                    ]
-                ),
-                np.array([[1.0, 1.0, 2.0, 2.0, 2, 0.8]])
-            ]
+            predictions = sv.Detections(...)
+            targets = sv.Detections(...)
 
-            mean_average_precison = sv.MeanAveragePrecision.from_tensors(
-                predictions=predictions,
-                targets=targets,
-            )
+            map_metric = MeanAveragePrecision()
+            map_result = map_metric.update(predictions, targets).compute()
 
-            print(mean_average_precison.map50_95)
-            # 0.6649
+            print(map_result)
+            print(map_result.map50_95)
+            map_result.plot()
             ```
         """
         (
@@ -243,6 +230,7 @@ def _compute(
             map50=map50,
             map75=map75,
             per_class_ap50_95=average_precisions,
+            metric_target=self._metric_target,
         )
 
     @staticmethod
@@ -256,7 +244,7 @@ def compute_average_precision(recall: np.ndarray, precision: np.ndarray) -> floa
             precision (np.ndarray): The precision curve.
 
         Returns:
-            float: Average precision.
+            (float): Average precision.
         """
         extended_recall = np.concatenate(([0.0], recall, [1.0]))
         extended_precision = np.concatenate(([1.0], precision, [0.0]))
@@ -320,7 +308,7 @@ def _average_precisions_per_class(
             eps (float, optional): Small value to prevent division by zero.
 
         Returns:
-            np.ndarray: Average precision for different IoU levels.
+            (np.ndarray): Average precision for different IoU levels.
         """
         eps = 1e-16
 
@@ -361,15 +349,44 @@ def _average_precisions_per_class(
 @dataclass
 class MeanAveragePrecisionResult:
     iou_thresholds: np.ndarray
+    """Array of IoU thresholds used in the calculations"""
     map50_95: float
+    """Mean Average Precision over IoU thresholds from 0.5 to 0.95"""
+
     map50: float
+    """Mean Average Precision at IoU threshold of 0.5"""
+
     map75: float
+    """Mean Average Precision at IoU threshold of 0.75"""
+
     per_class_ap50_95: np.ndarray
+    """Average precision for each class at different IoU thresholds"""
+
+    metric_target: MetricTarget
+    """
+    Defines the type of data used for the metric - boxes, masks or
+    oriented bounding boxes.
+    """
+
     small_objects: Optional[MeanAveragePrecisionResult] = None
+    """Mean Average Precision results for small objects"""
+
     medium_objects: Optional[MeanAveragePrecisionResult] = None
+    """Mean Average Precision results for medium objects"""
+
     large_objects: Optional[MeanAveragePrecisionResult] = None
+    """Mean Average Precision results for large objects"""
 
     def __str__(self) -> str:
+        """
+        Format the mAP results as a pretty string.
+
+        Example:
+            ```python
+            print(map_result)
+            ```
+        """
+
         out_str = (
             f"{self.__class__.__name__}:\n"
             f"iou_thresholds: {self.iou_thresholds}\n"
@@ -402,7 +419,7 @@ def to_pandas(self) -> "pd.DataFrame":
         Convert the result to a pandas DataFrame.
 
         Returns:
-            pd.DataFrame: The result as a DataFrame.
+            (pd.DataFrame): The result as a DataFrame.
         """
         ensure_pandas_installed()
         import pandas as pd