fix(pre_commit): 🎨 auto format pre-commit hooks

Author: pre-commit-ci[bot]

docs/learn/state-estimators.md

@@ -155,7 +155,6 @@ We can also benchmark the trackers using the different State Estimators and we g
 - With defaults parameters in Soccernet dataset, SORT tracker with XYXYStateEstimator has ~5% more HOTA than using XCYC,
 - In SportsMOT, for OC-SORT and ByteTrack, the StateEstimator doesn't affect the performance, while for SORT XYXYStateEstimator gives a small advantage of ~2% HOTA with default parameters and 0.4% when tuning both.
 
-
 But lets visualize where these differences are, here is an example where using XCYCSR State Estimator associates an occluded track correctly, while using XYXY changes the ID:
 
 <div style="display: flex; justify-content: center;">

trackers/core/botsort/tracker.py

@@ -72,6 +72,7 @@ class BoTSORTTracker(BaseTracker):
             int(frame_rate / 30.0 * lost_track_buffer)
             to maintain consistent “seconds” worth of buffer across different FPS.
     """
+
     def __init__(
         self,
         lost_track_buffer: int = 30,
@@ -179,10 +180,7 @@ def update(
         for track in self.tracks:
             if track.time_since_update > 1:
                 lost_tracks.append(track)
-            elif (
-                track.number_of_successful_updates
-                >= self.minimum_consecutive_frames
-            ):
+            elif track.number_of_successful_updates >= self.minimum_consecutive_frames:
                 confirmed_tracks.append(track)
             else:
                 unconfirmed_tracks.append(track)
@@ -208,8 +206,7 @@ def update(
             track = strack_pool[row]
             track.update(high_boxes[col])
             if (
-                track.number_of_successful_updates
-                >= self.minimum_consecutive_frames
+                track.number_of_successful_updates >= self.minimum_consecutive_frames
                 and track.tracker_id == -1
             ):
                 track.tracker_id = BoTSORTTracklet.get_next_tracker_id()
@@ -233,8 +230,7 @@ def update(
             track = remaining_tracked[row]
             track.update(low_boxes[col])
             if (
-                track.number_of_successful_updates
-                >= self.minimum_consecutive_frames
+                track.number_of_successful_updates >= self.minimum_consecutive_frames
                 and track.tracker_id == -1
             ):
                 track.tracker_id = BoTSORTTracklet.get_next_tracker_id()

trackers/core/botsort/tracklet.py

@@ -69,23 +69,36 @@ def _set_scale_aware_noise(
 
         Q = np.diag(
             [
-                (sp * w) ** 2, (sp * h) ** 2, (sp * w) ** 2, (sp * h) ** 2,
-                (sv * w) ** 2, (sv * h) ** 2, (sv * w) ** 2, (sv * h) ** 2,
+                (sp * w) ** 2,
+                (sp * h) ** 2,
+                (sp * w) ** 2,
+                (sp * h) ** 2,
+                (sv * w) ** 2,
+                (sv * h) ** 2,
+                (sv * w) ** 2,
+                (sv * h) ** 2,
             ]
         )
         R = np.diag(
             [
-                (sm * w) ** 2, (sm * h) ** 2, (sm * w) ** 2, (sm * h) ** 2,
+                (sm * w) ** 2,
+                (sm * h) ** 2,
+                (sm * w) ** 2,
+                (sm * h) ** 2,
             ]
         )
 
         if initial:
             P = np.diag(
                 [
-                    (2 * sp * w) ** 2, (2 * sp * h) ** 2,
-                    (2 * sp * w) ** 2, (2 * sp * h) ** 2,
-                    (10 * sv * w) ** 2, (10 * sv * h) ** 2,
-                    (10 * sv * w) ** 2, (10 * sv * h) ** 2,
+                    (2 * sp * w) ** 2,
+                    (2 * sp * h) ** 2,
+                    (2 * sp * w) ** 2,
+                    (2 * sp * h) ** 2,
+                    (10 * sv * w) ** 2,
+                    (10 * sv * h) ** 2,
+                    (10 * sv * w) ** 2,
+                    (10 * sv * h) ** 2,
                 ]
             )
             self.state_estimator.set_kf_covariances(R=R, Q=Q, P=P)
@@ -142,7 +155,6 @@ def get_state_bbox(self) -> np.ndarray:
         """Return the current bounding-box estimate in xyxy format."""
         return self.state_estimator.state_to_bbox()
 
-
     def apply_cmc(self, H: np.ndarray) -> None:
         """Apply a 2×3 affine camera-motion transform **in place**.
 
@@ -175,9 +187,7 @@ def apply_cmc(self, H: np.ndarray) -> None:
         kf.P = A @ kf.P @ A.T
 
     @staticmethod
-    def apply_cmc_batch(
-        tracklets: Sequence[BoTSORTTracklet], H: np.ndarray
-    ) -> None:
+    def apply_cmc_batch(tracklets: Sequence[BoTSORTTracklet], H: np.ndarray) -> None:
         """Apply a 2×3 affine camera-motion transform to all tracklets at once.
 
         Vectorised replacement for calling :meth:`apply_cmc` in a loop.
@@ -198,9 +208,7 @@ def apply_cmc_batch(
         dim = tracklets[0].state_estimator.kf.x.shape[0]
 
         # Stack states (N, dim) and covariances (N, dim, dim)
-        states = np.array(
-            [trk.state_estimator.kf.x.reshape(-1) for trk in tracklets]
-        )
+        states = np.array([trk.state_estimator.kf.x.reshape(-1) for trk in tracklets])
         Ps = np.array([trk.state_estimator.kf.P for trk in tracklets])
 
         # Batch-transform centre positions: x' = x @ R.T + t

trackers/core/botsort/utils.py

@@ -44,9 +44,7 @@ def get_alive_trackers(
     return alive_trackers
 
 
-def _fuse_score(
-    iou_similarity: np.ndarray, scores: np.ndarray
-) -> np.ndarray:
+def _fuse_score(iou_similarity: np.ndarray, scores: np.ndarray) -> np.ndarray:
     """Fuse IoU similarity matrix with detection confidence scores.
 
     Following the original ByteTrack implementation, the IoU similarity is