ZeroDivisionError: division by zero at covert to tracklets #1638

The videos are analyzed. Time to assemble animals and track 'em...
 Call 'create_video_with_all_detections' to check multi-animal detection quality before tracking.
If the tracking is not satisfactory for some videos, consider expanding the training set. You can use the function 'extract_outlier_frames' to extract a few representative outlier frames.
Using snapshot-200000 for model F:\thirdcolony\colony3new-shachaf-2021-05-25\dlc-models\iteration-0\colony3newMay25-trainset95shuffle5
Processing...  F:\thirdcolony\colony3new-shachaf-2021-05-25\videos\col3D1.mp4
Analyzing F:\thirdcolony\colony3new-shachaf-2021-05-25\videos\col3D1DLC_dlcrnetms5_colony3newMay25shuffle5_200000.h5
37it [00:00, 961.51it/s]
---------------------------------------------------------------------------
ZeroDivisionError                         Traceback (most recent call last)
~\.conda\envs\dlc 22\lib\site-packages\deeplabcut\gui\analyze_videos.py in convert2_tracklets(self, event)
    422             calibrate=self.calibrate.GetStringSelection() == "Yes",
    423             window_size=self.winsize.GetValue(),
--> 424             identity_only=self.identity_toggle.GetStringSelection() == "Yes",
    425         )
    426

~\.conda\envs\dlc 22\lib\site-packages\deeplabcut\pose_estimation_tensorflow\predict_videos.py in convert_detections2tracklets(config, videos, videotype, shuffle, trainingsetindex, overwrite, destfolder, ignore_bodyparts, inferencecfg, modelprefix, track_method, greedy, calibrate, window_size, identity_only)
   1461                     )
   1462                     ass.calibrate(train_data_file)
-> 1463                 ass.assemble()
   1464                 ass.to_pickle(dataname.split(".h5")[0] + "_assemblies.pickle")
   1465

~\.conda\envs\dlc 22\lib\site-packages\deeplabcut\pose_estimation_tensorflow\lib\inferenceutils.py in assemble(self, chunk_size, n_processes)
    751         if chunk_size == 0 or multiprocessing.get_start_method() == "spawn":
    752             for i, data_dict in enumerate(tqdm(self)):
--> 753                 assemblies, unique = self._assemble(data_dict, i)
    754                 if assemblies:
    755                     self.assemblies[i] = assemblies

~\.conda\envs\dlc 22\lib\site-packages\deeplabcut\pose_estimation_tensorflow\lib\inferenceutils.py in _assemble(self, data_dict, ind_frame)
    695             if self._kde is not None:
    696                 scores = [
--> 697                     -self.calc_assembly_mahalanobis_dist(ass) for ass in assemblies
    698                 ]
    699             else:

~\.conda\envs\dlc 22\lib\site-packages\deeplabcut\pose_estimation_tensorflow\lib\inferenceutils.py in <listcomp>(.0)
    695             if self._kde is not None:
    696                 scores = [
--> 697                     -self.calc_assembly_mahalanobis_dist(ass) for ass in assemblies
    698                 ]
    699             else:

~\.conda\envs\dlc 22\lib\site-packages\deeplabcut\pose_estimation_tensorflow\lib\inferenceutils.py in calc_assembly_mahalanobis_dist(self, assembly, return_proba, nan_policy)
    304             inv_cov = self._kde.inv_cov[np.ix_(inds, inds)]
    305             # Correct distance to account for missing observations
--> 306             factor = self._kde.d / len(inds)
    307         else:
    308             # Alternatively, reduce contribution of missing values to the Mahalanobis

ZeroDivisionError: division by zero