Update det-metrics.py

det-metrics.py

@@ -131,14 +131,18 @@ class DetectionMetric(evaluate.Metric):
         # save parameters for later
         self.payload = payload
         self.model_names = payload.models if payload else ["custom"]
+        self.iou_threshold = iou_threshold
+        self.area_ranges_tuples = area_ranges_tuples
+        self.class_agnostic = class_agnostic
+        self.iou_type = iou_type
+        self.bbox_format = bbox_format
+
+        # postprocess parameters
         self.iou_thresholds = (
             iou_threshold if isinstance(iou_threshold, list) else [iou_threshold]
         )
         self.area_ranges = [v for _, v in area_ranges_tuples]
         self.area_ranges_labels = [k for k, _ in area_ranges_tuples]
-        self.class_agnostic = class_agnostic
-        self.iou_type = iou_type
-        self.box_format = bbox_format
 
         # initialize coco_metrics
         self.coco_metric = PrecisionRecallF1Support(
@@ -147,7 +151,7 @@ class DetectionMetric(evaluate.Metric):
             area_ranges_labels=self.area_ranges_labels,
             class_agnostic=self.class_agnostic,
             iou_type=self.iou_type,
-            box_format=self.box_format,
+            box_format=self.bbox_format,
         )
 
         # initialize evaluation metric
@@ -248,7 +252,7 @@ class DetectionMetric(evaluate.Metric):
                 area_ranges_labels=self.area_ranges_labels,
                 class_agnostic=self.class_agnostic,
                 iou_type=self.iou_type,
-                box_format=self.box_format,
+                box_format=self.bbox_format,
             )
         return results
 
@@ -414,3 +418,75 @@ class DetectionMetric(evaluate.Metric):
         references = [{"boxes": [[1.0, 2.0, 3.0, 4.0]], "labels": [0], "area": [1.0]}]
 
         return predictions, references
+
+
+    def compute_from_payload(self, payload: Payload):
+        """
+        Compute the metric from the payload.
+        Args:
+            payload (Payload): The payload to compute the metric from.
+            **kwargs: Additional keyword arguments.
+        Returns:
+            dict: The computed metric results with the following format:
+            {
+              "model_name": {
+                "overall": {
+                  "all": {"tp": ..., "fp": ..., "fn": ..., "f1": ...},
+                  ...  # more area ranges
+                },
+                "per_sequence": {
+                  "sequence_name": {
+                    "all": {...},
+                    ...  # more area ranges
+                  },
+                  ...  # more sequences
+                }
+              },
+              ...  # more models
+            }
+        
+        Note:
+        - If the metric does not support area ranges, the metric should store the results under the `all` key.
+        - If a range area is provided it will be displayed in the output. if area_ranges_tuples is None, then all the area ranges will be displayed
+        """
+        results = {}
+
+        for model_name in payload.models:
+            results[model_name] = {"overall": {}, "per_sequence": {}}
+            
+            # per-sequence loop
+            for seq_name, sequence in payload.sequences.items():
+                # create new payload only with specific sequence and model
+                sequence_payload = Payload(
+                    dataset=payload.dataset,
+                    gt_field_name=payload.gt_field_name,
+                    models=[model_name],
+                    sequences={seq_name: sequence}
+                )
+                module = DetectionMetric(
+                    area_ranges_tuples=self.area_ranges_tuples,
+                    iou_threshold=self.iou_threshold,
+                    class_agnostic=self.class_agnostic,
+                    bbox_format=self.bbox_format,
+                    iou_type=self.iou_type,
+                    payload=sequence_payload
+                )
+                results[model_name]["per_sequence"][seq_name] = module.compute()[model_name]["metrics"]
+
+            # overall per-model loop
+            model_payload = Payload(
+                    dataset=payload.dataset,
+                    gt_field_name=payload.gt_field_name,
+                    models=[model_name],
+                    sequences=payload.sequences
+                )
+            module = DetectionMetric(
+                area_ranges_tuples=self.area_ranges_tuples,
+                iou_threshold=self.iou_threshold,
+                class_agnostic=self.class_agnostic,
+                bbox_format=self.bbox_format,
+                iou_type=self.iou_type,
+                payload=model_payload
+            )
+            results[model_name]["overall"] = module.compute()[model_name]["metrics"]
+        return results