Fixed bug in csv reader, which was shuffling bodyparts if they were not in a particular order

Pipfile

@@ -21,6 +21,7 @@ tensorflow-probability = "*"
 tensorflow = "*"
 tqdm = "*"
 umap-learn = "*"
+scikit-image = "*"
 
 [dev-packages]
 coverage = "*"

deepof/data.py

@@ -68,7 +68,7 @@ class project:
 
         self.table_format = table_format
         if self.table_format == "autodetect":
-            ex = os.listdir(self.table_path)[0]
+            ex = [i for i in os.listdir(self.table_path) if not i.startswith(".")][0]
             if ".h5" in ex:
                 self.table_format = ".h5"
             elif ".csv" in ex:
@@ -190,25 +190,15 @@ class project:
             }
         elif self.table_format == ".csv":
 
-            for tab in self.tables:
-                head = pd.read_csv(
-                    deepof.utils.os.path.join(self.table_path, tab), nrows=2
-                )
-                data = pd.read_csv(
+            tab_dict = {
+                deepof.utils.re.findall("(.*)DLC", tab)[0]: pd.read_csv(
                     deepof.utils.os.path.join(self.table_path, tab),
-                    skiprows=2,
-                    index_col="coords",
-                    dtype={"coords": int},
-                )
-                data.columns = pd.MultiIndex.from_product(
-                    [
-                        [head.columns[2]],
-                        set(list(head.iloc[0])[2:]),
-                        ["x", "y", "likelihood"],
-                    ],
-                    names=["scorer", "bodyparts", "coords"],
+                    header=[0, 1, 2],
+                    index_col=0,
+                    dtype=float,
                 )
-                tab_dict[deepof.utils.re.findall("(.*)DLC", tab)[0]] = data
+                for tab in self.tables
+            }
 
         lik_dict = defaultdict()
 

deepof/pose_utils.py

@@ -376,7 +376,7 @@ def get_hparameters(hparams: dict = {}) -> dict:
         - hparams (dict): dictionary containing hyperparameters to overwrite
 
     Returns:
-        - defaults (dict): dictionary with overwriten parameters. Those not
+        - defaults (dict): dictionary with overwritten parameters. Those not
         specified in the input retain their default values"""
 
     defaults = {
@@ -548,7 +548,7 @@ def rule_based_tagging(
 
     for _id in animal_ids:
         tag_dict[_id + undercond + "climbing"] = deepof.utils.smooth_boolean_array(
-            climb_wall(arena_type, arena, coords, w / 100, _id + undercond + "Nose")
+            climb_wall(arena_type, arena, coords, 1, _id + undercond + "Nose")
         )
         tag_dict[_id + undercond + "speed"] = speeds[_id + undercond + "Center"]
         tag_dict[_id + undercond + "huddle"] = deepof.utils.smooth_boolean_array(
@@ -578,11 +578,13 @@ def tag_rulebased_frames(
     dims,
     undercond,
     hparams,
+    arena,
 ):
     """Helper function for rule_based_video. Annotates a given frame with on-screen information
     about the recognised patterns"""
 
     w, h = dims
+    arena, h, w = arena
 
     def write_on_frame(text, pos, col=(255, 255, 255)):
         """Partial closure over cv2.putText to avoid code repetition"""
@@ -614,7 +616,7 @@ def tag_rulebased_frames(
 
     if len(animal_ids) > 1:
 
-        cv2.line(frame, (100, 100), (110, 100), (255, 0, 0))
+        cv2.circle(frame, (arena[0], arena[1]), arena[2], thickness=2, color=(0, 0, 255))
 
         if tag_dict["nose2nose"][fnum] and not tag_dict["sidebyside"][fnum]:
             write_on_frame("Nose-Nose", conditional_pos())
@@ -775,6 +777,7 @@ def rule_based_video(
             (w, h),
             undercond,
             hparams,
+            (arena, h, w)
         )
 
         if writer is None:

deepof/utils.py

@@ -19,6 +19,7 @@ import regex as re
 from copy import deepcopy
 from itertools import combinations, product
 from joblib import Parallel, delayed
+#from skimage.transform import hough_ellipse
 from sklearn import mixture
 from tqdm import tqdm
 from typing import Tuple, Any, List, Union, NewType
@@ -410,16 +411,21 @@ def circular_arena_recognition(frame: np.array) -> np.array:
     gray_image = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
     ret, thresh = cv2.threshold(gray_image, 50, 255, 0)
     frame = cv2.medianBlur(thresh, 9)
-    circle = cv2.HoughCircles(
+    circle = hough_ellipse(
         frame,
-        cv2.HOUGH_GRADIENT,
-        1,
-        300,
-        param1=50,
-        param2=10,
-        minRadius=0,
-        maxRadius=0,
+        # accuracy=20,
+        # threshold=250,
+        # min_size=frame.shape[0] // 6,
+        # max_size=frame.shape[0] // 2,
+       cv2.HOUGH_GRADIENT,
+       1,
+       300,
+       param1=50,
+       param2=10,
+       minRadius=0,
+       maxRadius=frame.shape[0] // 2,
     )
+    result.sort(order='accumulator')
 
     circles = []