Python crashes for holistic model of mediapipe if I am applying the model to a video.
I tried different steps to solve that issue:

1. Using mediapipe in MacOs and Windows
2. Use of CPU and GPU delegates
3. Use of images instead of videos
4. Conversion of the original MOV to MP4 format
   However, this did not resolve the issue. I wonder whether this issue is related to the file format (MP4/MOV), but cv2 is able to read that and plotting the numpy array results in the expected image.

Thus, I would highly appreciate your support. You can find an example code attached:

My code:

import cv2
import sys
import os
import glob

import numpy as np
import mediapipe as mp  # mediapipe for hand dectection

import plotly.graph_objs as go
from plotly.subplots import make_subplots

from matplotlib import pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
import matplotlib.animation

import pandas as pd
import plotly.graph_objects as go
import plotly.io as pio
import scipy as sc
from scipy import signal
from mediapipe.tasks import python


BaseOptions =python.BaseOptions
HolisticLandmarker = mp.tasks.vision.HolisticLandmarker
HolisticLandmarkerOptions = mp.tasks.vision.HolisticLandmarkerOptions
VisionRunningMode = mp.tasks.vision.RunningMode

# Create a face landmarker instance with the video mode:
options = HolisticLandmarkerOptions(
    base_options=BaseOptions(model_asset_path="/data/Christian/MoCap/holistic_landmarker.task"),
    min_face_detection_confidence=0.8, 
    min_face_suppression_threshold=0.8, 
    min_face_landmarks_confidence=0.8, 
    min_pose_detection_confidence=0.8, 
    min_pose_suppression_threshold=0.8, 
    min_pose_landmarks_confidence=0.8, 
    min_hand_landmarks_confidence=0.8, 
    output_face_blendshapes=True,
    running_mode=VisionRunningMode.VIDEO)


input_dir = r'/data/video' 

video_files = np.array(glob.glob(os.path.join(input_dir, "**",'*mp4'),recursive=True))
# Set mediapipe model

# Iterate through video_file (words)
for video_file in video_files:

    # Create a directory for the current word
    video_file_path = video_file[:video_file.rfind(os.sep)]
    outputName = video_file[video_file.rfind(
        os.sep)+1:video_file.rfind('video')]
    output_video_path = os.path.join(
        output_dir, video_file[video_file.rfind("/")+1:-4])
    subject = outputName[outputName.rfind(
        "sub-"):outputName.rfind("_ses")]
    session = outputName[outputName.rfind(
        "ses-"):outputName.find("_", outputName.find("_")+1)]

    if not os.path.exists(os.path.join(output_video_path, subject, session)):
        os.makedirs(os.path.join(output_video_path, subject,
                    session, '.video'), exist_ok=True)
        os.makedirs(os.path.join(output_video_path, subject,
                    session, 'motion'), exist_ok=True)

    video_path = os.path.join(video_file_path, video_file)
    
    with HolisticLandmarker.create_from_options(options) as holistic:
        # The landmarker is initialized. Use it here.
        # Read video and calculate duration
        cap = cv2.VideoCapture(video_path)

        # Saving Video-
        fps = int(cap.get(cv2.CAP_PROP_FPS))
        fourcc = cv2.VideoWriter_fourcc(*'mp4v')
        out = cv2.VideoWriter(os.path.join(output_video_path, subject, session, '.video', outputName +
                              'desc-overlayVideo_video.mov'), fourcc, fps, (int(cap.get(3)), int(cap.get(4))))
        # Initialize empty list for keypoints
        keypoints_list = []
        success = 1
        # Iterate through frames in the sequence
        while success:
            # Read video frame
            success, frame = cap.read()
            frame_timestamp_ms = int(cap.get(cv2.CAP_PROP_POS_MSEC))

            # Make detections and draw landmarks
            mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=frame)
            results = holistic.detect_for_video(mp_image, frame_timestamp_ms)

Here is the resulting error message:

I0000 00:00:1717520611.283032   68206 gl_context_egl.cc:85] Successfully initialized EGL. Major : 1 Minor: 5
I0000 00:00:1717520611.285679   70364 gl_context.cc:357] GL version: 3.2 (OpenGL ES 3.2 Mesa 23.2.1-1ubuntu3.1~22.04.2), renderer: llvmpipe (LLVM 15.0.7, 256 bits)


Fatal Python error: Aborted


Main thread:
Current thread 0x00007f0e2b8eb440 (most recent call first):
  File "/home/christian.schmitz/.conda/envs/MNI-mocap/lib/python3.8/site-packages/mediapipe/python/packet_getter.py", line 61 in get_proto
  File "/home/christian.schmitz/.conda/envs/MNI-mocap/lib/python3.8/site-packages/mediapipe/tasks/python/vision/holistic_landmarker.py", line 170 in _build_landmarker_result
  File "/home/christian.schmitz/.conda/envs/MNI-mocap/lib/python3.8/site-packages/mediapipe/tasks/python/vision/holistic_landmarker.py", line 535 in detect_for_video
  File "/tmp/ipykernel_68206/2139099985.py", line 103 in <module>


Restarting kernel...