Implemented outlier interpolation

deepof/data.py

@@ -63,7 +63,7 @@ class project:
         interpolate_outliers: bool = True,
         interpolation_limit: int = 5,
         interpolation_std: int = 5,
-        likelihood_tol: float = 0.75,
+        likelihood_tol: float = 0.25,
         model: str = "mouse_topview",
         path: str = deepof.utils.os.path.join("."),
         smooth_alpha: float = 0.99,
@@ -296,7 +296,9 @@ class project:
                 print("Iterative imputation of ocluded bodyparts...")
 
             for k, value in tab_dict.items():
-                imputed = IterativeImputer(max_iter=250, skip_complete=True).fit_transform(value)
+                imputed = IterativeImputer(
+                    max_iter=250, skip_complete=True
+                ).fit_transform(value)
                 tab_dict[k] = pd.DataFrame(
                     imputed, index=value.index, columns=value.columns
                 )

deepof/pose_utils.py

@@ -652,27 +652,25 @@ def tag_rulebased_frames(
         )
     )
 
-    if len(animal_ids) > 1:
+    if debug:
+        # Print arena for debugging
+        cv2.ellipse(frame, arena[0], arena[1], arena[2], 0, 360, (0, 255, 0), 3)
+        # Print body parts for debuging
+        for bpart in coords.columns.levels[0]:
+            if not np.isnan(coords[bpart]["x"][fnum]):
+                cv2.circle(
+                    frame,
+                    (int(coords[bpart]["x"][fnum]), int(coords[bpart]["y"][fnum])),
+                    radius=3,
+                    color=(
+                        (255, 0, 0) if bpart.startswith(animal_ids[0]) else (0, 0, 255)
+                    ),
+                    thickness=-1,
+                )
+        # Print frame number
+        write_on_frame("Frame " + str(fnum), (int(w * 0.3 / 10), int(h / 1.15)))
 
-        if debug:
-            # Print arena for debugging
-            cv2.ellipse(frame, arena[0], arena[1], arena[2], 0, 360, (0, 255, 0), 3)
-            # Print body parts for debuging
-            for bpart in coords.columns.levels[0]:
-                if not np.isnan(coords[bpart]["x"][fnum]):
-                    cv2.circle(
-                        frame,
-                        (int(coords[bpart]["x"][fnum]), int(coords[bpart]["y"][fnum])),
-                        radius=3,
-                        color=(
-                            (255, 0, 0)
-                            if bpart.startswith(animal_ids[0])
-                            else (0, 0, 255)
-                        ),
-                        thickness=-1,
-                    )
-            # Print frame number
-            write_on_frame("Frame " + str(fnum), (int(w * 0.3 / 10), int(h / 1.15)))
+    if len(animal_ids) > 1:
 
         if tag_dict["nose2nose"][fnum] and not tag_dict["sidebyside"][fnum]:
             write_on_frame("Nose-Nose", conditional_pos())
@@ -721,7 +719,7 @@ def tag_rulebased_frames(
         if len(animal_ids) > 1:
             colcond = frame_speeds[_id] == max(list(frame_speeds.values()))
         else:
-            colcond = hparams["huddle_speed"] > frame_speeds
+            colcond = hparams["huddle_speed"] < frame_speeds
 
         write_on_frame(
             str(

deepof/utils.py

@@ -460,7 +460,7 @@ def interpolate_outliers(
     lag: int = 5,
     n_std: int = 3,
     mode: str = "or",
-    limit: int = 15,
+    limit: int = 10,
 ):
     """Marks all outliers in experiment and replaces them using a univariate linear interpolation approach.
     Note that this approach only works for equally spaced data (constant camera acquisition rates).