Not working on Windows 11: mediapipe version 0.10.18 and 0.10.20

[creativival]

I have experimented with pip installation of MediaPipe to Windows 11.

1. factory reset the laptop and boot Pure Windows 11
2. install Python 3.12.8 from the Python official website
3. install MediaPipe by "pip install mediapipe==0.10.14"
4. run the app and get error “Import Error: DLL load failed while importing _framework_bindings"
5. pip install msvc-runtime
6. re-run the app and it launches normally

Now, when I upgrade MediaPipe, I get the same error.

1. pip uninstall mediapipe
2. pip install mediapipe=0.10.18
3. when I start the app, I get the error “Import Error: DLL load failed while importing _framework_bindings”
4. pip uninstall mediapipe
5. pip install mediapipe== 0.10.20
6. When you start the app, the error “Import Error: DLL load failed while importing _framework_bindings”
7. pip uninstall mediapipe
8. pip install mediapipe=0.10.14
9. when the application is launched, it launches normally.

Based on the above verification, the new version cannot be launched on Windows 11.
In addition, the same problem occurs after installing the Visual C++ redistributable package.

Has anyone successfully installed and launched in a Windows 11 environment?　Please let me know if there is a workaround.

[kuaashish]

Hi @creativival,

We attempted to reproduce the issue by installing MediaPipe versions 0.10.14, 0.10.18, and 0.10.20 on Windows 11 with Python 3.12.8. While the installation was successful, we encountered the same error as you when attempting to import MediaPipe.

However, after installing msvc-runtime using pip install msvc-runtime, we were able to successfully import MediaPipe.

Could you please try installing msvc-runtime and check if the issue is resolved? If the problem persists, kindly provide the complete error logs along with your Windows version details.

Thank you!!

[creativival]

Hi @kuaashish

Thanks for your reply.

As you said, “import mediapipe” is possible in the latest version (0.10.20), but the following Pose Landmarks sample code gives an error “Import Error: DLL load failed while importing _framework_bindings".

It worked fine in the older version (0.10.14).

Please try the following code with “image.jpg” and “pose_landmarker.task” in the same directory.

I installed the following libraries: "pip install mediapipe opencv-python msvc-runtime"

My environment: Windows 11 Pro 23H2 + Python 3.12.8 (Official) + VS Code

Thank you!!

Sample code is a modified version of the official python sample code for colab to work locally.

import mediapipe as mp
from mediapipe.tasks import python
from mediapipe.tasks.python import vision
from mediapipe.framework.formats import landmark_pb2
import numpy as np
import cv2


def draw_landmarks_on_image(rgb_image, detection_result):
  # RGB → BGRに変換
  bgr_image = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2BGR)
  annotated_image = np.copy(bgr_image)

  pose_landmarks_list = detection_result.pose_landmarks

  # Loop through the detected poses to visualize.
  for idx in range(len(pose_landmarks_list)):
    pose_landmarks = pose_landmarks_list[idx]

    # Draw the pose landmarks.
    pose_landmarks_proto = landmark_pb2.NormalizedLandmarkList()
    pose_landmarks_proto.landmark.extend([
      landmark_pb2.NormalizedLandmark(x=landmark.x, y=landmark.y, z=landmark.z) for landmark in pose_landmarks
    ])
    mp.solutions.drawing_utils.draw_landmarks(
      annotated_image,
      pose_landmarks_proto,
      mp.solutions.pose.POSE_CONNECTIONS,
      mp.solutions.drawing_styles.get_default_pose_landmarks_style())

  # BGR → RGBに変換
  annotated_image_rgb = cv2.cvtColor(annotated_image, cv2.COLOR_BGR2RGB)
  return annotated_image_rgb

# STEP 2: Create an PoseLandmarker object.
# base_options = python.BaseOptions(model_asset_path='pose_landmarker.task')  # macOS
base_options = python.BaseOptions(model_asset_buffer=open('pose_landmarker.task', "rb").read())  # Windows
options = vision.PoseLandmarkerOptions(
    base_options=base_options,
    output_segmentation_masks=True)
detector = vision.PoseLandmarker.create_from_options(options)

# STEP 3: Load the input image.
image = mp.Image.create_from_file("image.jpg")

# STEP 4: Detect pose landmarks from the input image.
detection_result = detector.detect(image)

# STEP 5: Process the detection result. In this case, visualize it.
annotated_image = draw_landmarks_on_image(image.numpy_view(), detection_result)
cv2.imshow('', cv2.cvtColor(annotated_image, cv2.COLOR_RGB2BGR))
cv2.waitKey(0)
cv2.destroyAllWindows()

[kuaashish]

Hi @creativival,

Thank you for sharing the modified code. However, we are still unable to reproduce the issue. The sample code is working fine in my VS Code environment when using image.jpg and the task file in the same directory and this code.

import mediapipe as mp
from mediapipe.tasks import python
from mediapipe.tasks.python import vision
from mediapipe.framework.formats import landmark_pb2
import numpy as np
import cv2


def draw_landmarks_on_image(rgb_image, detection_result):
  # RGB → BGRに変換
  bgr_image = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2BGR)
  annotated_image = np.copy(bgr_image)

  pose_landmarks_list = detection_result.pose_landmarks

  # Loop through the detected poses to visualize.
  for idx in range(len(pose_landmarks_list)):
    pose_landmarks = pose_landmarks_list[idx]

    # Draw the pose landmarks.
    pose_landmarks_proto = landmark_pb2.NormalizedLandmarkList()
    pose_landmarks_proto.landmark.extend([
      landmark_pb2.NormalizedLandmark(x=landmark.x, y=landmark.y, z=landmark.z) for landmark in pose_landmarks
    ])
    mp.solutions.drawing_utils.draw_landmarks(
      annotated_image,
      pose_landmarks_proto,
      mp.solutions.pose.POSE_CONNECTIONS,
      mp.solutions.drawing_styles.get_default_pose_landmarks_style())

  # BGR → RGBに変換
  annotated_image_rgb = cv2.cvtColor(annotated_image, cv2.COLOR_BGR2RGB)
  return annotated_image_rgb

# STEP 2: Create an PoseLandmarker object.
# base_options = python.BaseOptions(model_asset_path='pose_landmarker.task')  # macOS
base_options = python.BaseOptions(model_asset_buffer=open('C:\Users\kuaashish\Downloads\pose_landmarker_full.task', "rb").read())  # Windows
options = vision.PoseLandmarkerOptions(
    base_options=base_options,
    output_segmentation_masks=True)
detector = vision.PoseLandmarker.create_from_options(options)

# STEP 3: Load the input image.
image = mp.Image.create_from_file("C:\Users\kuaashish\Downloads\Pose.jpg")

# STEP 4: Detect pose landmarks from the input image.
detection_result = detector.detect(image)

# STEP 5: Process the detection result. In this case, visualize it.
annotated_image = draw_landmarks_on_image(image.numpy_view(), detection_result)
cv2.imshow('', cv2.cvtColor(annotated_image, cv2.COLOR_RGB2BGR))
cv2.waitKey(0)
cv2.destroyAllWindows()

As shown in the attached screenshot

Could you try creating a new Python environment and reinstalling MediaPipe from scratch?

Thank you!!

[creativival]

Hi @kuaashish

I found the cause of the bug.

When reinstalling Windows 11, if I choose “Local reinstall”, I get the error “Import Error: DLL load failed while importing _framework_bindings” when I run the above Issue.py
If I choose “Cloud download”, it works fine.

Another Isuue

There is another issue that may be related to this: the same error occurs in the import order with the Panda3D library.

Steps to reproduce

Install the library with “pip install mediapipe panda3d”.
Run interactive mode

>> from direct.showbase.ShowBase import ShowBase
>> import mediapipe

Error “Import Error: DLL load failed while importing _framework_bindings

This occurs with “0.10.18” and “0.10.20” and not with “0.10.14”. Please verify.

[kuaashish]

Hi @creativival,

We have identified another issue and will bring it to our team's attention. However, if you previously had no issues importing MediaPipe and later imported other dependencies, you can avoid the error by using versions 0.10.18 or 0.10.20, as shown below.

Thank you!!