perf: weighted score for best label/score

biotrack/track.py

@@ -1,8 +1,7 @@
 # biotrack, Apache-2.0 license
 # Filename: biotrack/track.py
 # Description:  Basic track object to contain and update tracks
-from collections import Counter
-
+from collections import defaultdict
 from biotrack.logger import info, debug
 import numpy as np
 
@@ -90,6 +89,10 @@ def rescale(self, pt: np.array, box: np.array) -> (np.array, np.array):
             box_rescale = box
         return pt_rescale, box_rescale
 
+    @property
+    def num_frames(self):
+        return self.last_updated_frame - self.start_frame + 1
+
     def get_best(self, rescale=True) -> (int, np.array, str, np.array, float):
         # Get the best box which is a few frames behind the last_updated_frame
         # This is pretty arbitrary, but sometimes the last box is too blurry or not visible
@@ -159,11 +162,16 @@ def update(self, label: str, pt: np.array, emb: np.array, frame_num: int, box:np
             scores = scores[-10:]
             labels = labels[-10:]
 
-        # Update the best_label with that of the highest scoring recent label
-        max_score = max(scores)
-        max_frame = np.argmax(scores)
-        self.best_label = labels[max_frame]
-        self.best_score = max_score
+        # Calculate the weighted score for each label
+        label_scores = defaultdict(float)
+        for score, label in zip(scores, labels):
+            label_scores[label] += float(score)
+
+        # Update the best_label with a weighted probability
+        label = max(label_scores, key=label_scores.get)
+        num_best_label = labels.count(label)
+        self.best_label = label
+        self.best_score = label_scores[label] / num_best_label
 
         pts_pretty = [f"{pt[0]:.2f},{pt[1]:.2f},{label},{score}" for pt, label, score in zip(self.pt.values(), self.label.values(), self.score.values())]
         total_frames = len(self.pt)