make wandb functionality also usable for class specific setting

det-metrics.py

@@ -29,10 +29,10 @@ LABEL_MAPPING = {
     'FISHING_SHIP': 0,
     'CONTAINER_SHIP': 0,
     'CRUISE_SHIP': 0,
+    'BOAT': 1,
     'BOAT_WITHOUT_SAILS': 1,
     'MOTORBOAT': 1,
     'MARITIME_VEHICLE': 1,
-    'BOAT': 1,
     'SAILING_BOAT': 2, 
     'SAILING_BOAT_WITH_CLOSED_SAILS': 2,
     'SAILING_BOAT_WITH_OPEN_SAILS': 2,
@@ -288,7 +288,6 @@ class DetectionMetric(evaluate.Metric):
             # add payload if available (otherwise predictions and references must be added with add function)
             if self.payload:
                 self._add_payload(self.payload, model_name)
-
             results[model_name] = self.coco_metric.compute()
 
             # reset coco_metrics for next model
@@ -299,6 +298,7 @@ class DetectionMetric(evaluate.Metric):
                 class_agnostic=self.class_agnostic,
                 iou_type=self.iou_type,
                 box_format=self.bbox_format,
+                labels=sorted(list(set(list(self.label_mapping.values())))) if self.label_mapping else None
             )
         return results
 
@@ -428,11 +428,6 @@ class DetectionMetric(evaluate.Metric):
             wandb: To interact with the Weights and Biases platform.
             datetime: To generate a timestamp for run names.
         """
-        if not self.class_agnostic:
-            raise ValueError(
-                "This method is not yet implemented for `self.class_agnostic=False`."
-            )
-        
         import os
         import wandb
         import datetime
@@ -449,7 +444,25 @@ class DetectionMetric(evaluate.Metric):
                 run = wandb.init(project=wandb_project, name=f"{k}-{formatted_datetime}")
             else:
                 run = wandb_runs[i]
-            run.log({f"{wandb_section}/{m}" : v for m, v in results[k]['metrics'].items()} if wandb_section is not None else results[k]['metrics'])
+            if self.class_agnostic:
+                run.log({f"{wandb_section}/{m}" : v for m, v in results[k]['metrics'].items()} if wandb_section is not None else results[k]['metrics'])
+            else:
+                for area_range, area_v in results[k]['metrics'].items():
+                    for m, v in area_v.items():
+                        if m in ["precision", "recall", "f1", "duplicates", "support"]:
+                            for cls_idx in list(set(list(self.label_mapping.values()))):
+                                cls_name = list(self.label_mapping.keys())[list(self.label_mapping.values()).index(cls_idx)]
+                                run.log(
+                                    {
+                                        (f"{(wandb_section + '/') if wandb_section else ''}{cls_name}/{area_range + '.' + m}") : v[cls_idx]
+                                    }
+                                )
+                        else:
+                            run.log(
+                                {
+                                    f"{(wandb_section + '/') if wandb_section is not None else ''}{m}" : v
+                                }
+                            )
             if wandb_runs is None: 
                 run.finish()