If tensorflow is imported first, media pipe solution hangs indefinitely

[NikhilBadami]

I was recently using the pose estimation solution in mediapipe in Python and came across an interesting bug. I found that if I imported tensorflow before importing mediapipe, it would cause the mediapipe library to hang indefinitely, but importing tensorflow second fixed this issue. The code I used to test this is below:

# import tensorflow --> Uncommenting this line will cause this file to hang
import cv2
import mediapipe as mp
mp_drawing = mp.solutions.drawing_utils
mp_pose = mp.solutions.pose
if __name__ == "__main__":
    # For static images:
    pose = mp_pose.Pose(static_image_mode=True, min_detection_confidence=0.5)
    path = "some/image/path"
    image = cv2.imread(path)
    image_hight, image_width, _ = image.shape
    # Convert the BGR image to RGB before processing.
    results = pose.process(cv2.cvtColor(image, cv2.COLOR_BGR2RGB)) --> This is the line the script hangs on

    # Draw pose landmarks on the image.
    annotated_image = image.copy()
    mp_drawing.draw_landmarks(
      annotated_image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)
    cv2.imshow('', annotated_image)
    cv2.waitKey(0)
    cv2.destroyAllWindows()
    pose.close()

I am working on a Mac on Catalina

[jiuqiant]

It hangs at https://github.com/google/mediapipe/blob/master/mediapipe/python/solution_base.py#L301 where it's waiting for the results coming back from the c++ graph execution logic.

MediaPipe has the c++ tflite inference engine in its binary. So, my initial guess is that the weird behavior is caused by the conflicts between MediaPipe's builtin tflite inference engine and TensorFlow's tflite inference engine. It's possible that who gets imported first will be registered and the latter one will be totally ignored.

[danhd123]

Same thing happens with ObjC/Swift on iOS. Mediapipe should probably include a warning in the documentation that you can't import Tensorflow separately and expect it to work.

[4sfaloth]

any news on this? I may be experiencing the same issue, though it is not fixed by importing tensorflow first

[google-ml-butler]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

[google-ml-butler]

Closing as stale. Please reopen if you'd like to work on this further.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No

[sgowroji]

We are reopening this issue as this still exist even with the latest release.

[flacle]

It hangs at https://github.com/google/mediapipe/blob/master/mediapipe/python/solution_base.py#L301 where it's waiting for the results coming back from the c++ graph execution logic.

MediaPipe has the c++ tflite inference engine in its binary. So, my initial guess is that the weird behavior is caused by the conflicts between MediaPipe's builtin tflite inference engine and TensorFlow's tflite inference engine. It's possible that who gets imported first will be registered and the latter one will be totally ignored.

In case you are working with classes, for this special case it might help to scope tensorflow on a class level by using import_module from importlib. Something like this in your __init__():

self.__tf = importlib.import_module('tensorflow')

[sureshdagooglecom]

Hi @NikhilBadami ,
Could you please try in latest version and confirm if working fine .

[google-ml-butler]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

[google-ml-butler]

Closing as stale. Please reopen if you'd like to work on this further.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No