feat: ⚡️ Incorporaten validation related changes from rhoadesj/dev

dacapo/experiments/tasks/evaluators/instance_evaluator.py

@@ -3,22 +3,48 @@
 from .evaluator import Evaluator
 from .instance_evaluation_scores import InstanceEvaluationScores
 
-from funlib.evaluate import rand_voi
+from funlib.evaluate import rand_voi, detection_scores
+
+try:
+    from funlib.segment.arrays import relabel
+
+    iou = True
+except ImportError:
+    iou = False
 
 import numpy as np
 
 
 class InstanceEvaluator(Evaluator):
-    criteria = ["voi_merge", "voi_split", "voi"]
+    criteria = ["voi_merge", "voi_split", "voi", "avg_iou"]
 
     def evaluate(self, output_array_identifier, evaluation_array):
         output_array = ZarrArray.open_from_array_identifier(output_array_identifier)
         evaluation_data = evaluation_array[evaluation_array.roi].astype(np.uint64)
         output_data = output_array[output_array.roi].astype(np.uint64)
         results = rand_voi(evaluation_data, output_data)
+        if iou:
+            try:
+                output_data, _ = relabel(output_data)
+                results.update(
+                    detection_scores(
+                        evaluation_data,
+                        output_data,
+                        matching_score="iou",
+                    )
+                )
+            except Exception:
+                results["avg_iou"] = 0
+                logger.warning(
+                    "Could not compute IoU because of an unknown error. Sorry about that."
+                )
+        else:
+            results["avg_iou"] = 0
 
         return InstanceEvaluationScores(
-            voi_merge=results["voi_merge"], voi_split=results["voi_split"]
+            voi_merge=results["voi_merge"],
+            voi_split=results["voi_split"],
+            avg_iou=results["avg_iou"],
         )
 
     @property

dacapo/experiments/validation_scores.py

@@ -113,7 +113,7 @@ def get_best(
         best value in two seperate arrays.
         """
         if "criteria" in data.coords.keys():
-            if len(data.coords["criteria"].shape) == 1:
+            if len(data.coords["criteria"].shape) > 1:
                 criteria_bests: List[Tuple[xr.DataArray, xr.DataArray]] = []
                 for criterion in data.coords["criteria"].values:
                     if self.evaluation_scores.higher_is_better(criterion.item()):
@@ -142,7 +142,10 @@ def get_best(
                 return (da_best_indexes, da_best_scores)
             else:
                 if self.evaluation_scores.higher_is_better(
-                    data.coords["criteria"].item()
+                    list(data.coords["criteria"].values)[
+                        0
+                    ]  # TODO: what is the intended behavior here? (hot fix in place)
+                    # data.coords["criteria"].item()
                 ):
                     return (
                         data.idxmax(dim, skipna=True, fill_value=None),