Visual Studio Code - ImportError: DLL load failed while importing onnx_cpp2py_export: A dynamic link library (DLL) initialization routine failed

VishIsHere26
October 20, 2024
22 views
1 vote
2 Answers

I’m trying to set up and test the OCR model from the doctr library provided by Mindee
(https://github.com/mindee/doctr)
within a VSCode. I installed the package in run.ipynb using the following command:

!pip install -qe doctr/.

I then tried to run the sample OCR code provided in the repository:

from doctr.io import DocumentFile
from doctr.models import ocr_predictor

However, I encountered the following error:

    "name": "ImportError",
    "message": "DLL load failed while importing onnx_cpp2py_export: A dynamic link library (DLL) initialization routine failed.",
    "stack": "---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
Cell In[3], line 2
      1 from doctr.io import DocumentFile
----> 2 from doctr.models import ocr_predictor
      4 model = ocr_predictor(pretrained=True)
      5 # PDF

File c:\Users\vishw\Desktop\test\doctr\doctr\__init__.py:1
----> 1 from . import io, models, datasets, contrib, transforms, utils
      2 from .file_utils import is_tf_available, is_torch_available
      3 from .version import __version__  # noqa: F401

File c:\Users\vishw\Desktop\test\doctr\doctr\models\__init__.py:1
----> 1 from .classification import *
      2 from .detection import *
      3 from .recognition import *

File c:\Users\vishw\Desktop\test\doctr\doctr\models\classification\__init__.py:1
----> 1 from .mobilenet import *
      2 from .resnet import *
      3 from .vgg import *

File c:\Users\vishw\Desktop\test\doctr\doctr\models\classification\mobilenet\__init__.py:4
      1 from doctr.file_utils import is_tf_available, is_torch_available
      3 if is_tf_available():
----> 4     from .tensorflow import *
      5 elif is_torch_available():
      6     from .pytorch import *

File c:\Users\vishw\Desktop\test\doctr\doctr\models\classification\mobilenet\tensorflow.py:15
     12 from tensorflow.keras import layers
     13 from tensorflow.keras.models import Sequential
---> 15 from ....datasets import VOCABS
     16 from ...utils import conv_sequence, load_pretrained_params
     18 __all__ = [
     19     "MobileNetV3",
     20     "mobilenet_v3_small",
   (...)
     25     "mobilenet_v3_small_page_orientation",
     26 ]

File c:\Users\vishw\Desktop\test\doctr\doctr\datasets\__init__.py:3
      1 from doctr.file_utils import is_tf_available
----> 3 from .generator import *
      4 from .cord import *
      5 from .detection import *

File c:\Users\vishw\Desktop\test\doctr\doctr\datasets\generator\__init__.py:4
      1 from doctr.file_utils import is_tf_available, is_torch_available
      3 if is_tf_available():
----> 4     from .tensorflow import *
      5 elif is_torch_available():
      6     from .pytorch import *  # type: ignore[assignment]

File c:\Users\vishw\Desktop\test\doctr\doctr\datasets\generator\tensorflow.py:8
      1 # Copyright (C) 2021-2024, Mindee.
      2 
      3 # This program is licensed under the Apache License 2.0.
      4 # See LICENSE or go to <https://opensource.org/licenses/Apache-2.0> for full license details.
      6 import tensorflow as tf
----> 8 from .base import _CharacterGenerator, _WordGenerator
     10 __all__ = ["CharacterGenerator", "WordGenerator"]
     13 class CharacterGenerator(_CharacterGenerator):

File c:\Users\vishw\Desktop\test\doctr\doctr\datasets\generator\base.py:14
     11 from doctr.io.image import tensor_from_pil
     12 from doctr.utils.fonts import get_font
---> 14 from ..datasets import AbstractDataset
     17 def synthesize_text_img(
     18     text: str,
     19     font_size: int = 32,
   (...)
     22     text_color: Optional[Tuple[int, int, int]] = None,
     23 ) -> Image.Image:
     24     """Generate a synthetic text image
     25 
     26     Args:
   (...)
     36         PIL image of the text
     37     """

File c:\Users\vishw\Desktop\test\doctr\doctr\datasets\datasets\__init__.py:4
      1 from doctr.file_utils import is_tf_available, is_torch_available
      3 if is_tf_available():
----> 4     from .tensorflow import *
      5 elif is_torch_available():
      6     from .pytorch import *  # type: ignore[assignment]

File c:\Users\vishw\Desktop\test\doctr\doctr\datasets\datasets\tensorflow.py:15
     11 import tensorflow as tf
     13 from doctr.io import read_img_as_tensor, tensor_from_numpy
---> 15 from .base import _AbstractDataset, _VisionDataset
     17 __all__ = ["AbstractDataset", "VisionDataset"]
     20 class AbstractDataset(_AbstractDataset):

File c:\Users\vishw\Desktop\test\doctr\doctr\datasets\datasets\base.py:16
     13 from doctr.io.image import get_img_shape
     14 from doctr.utils.data import download_from_url
---> 16 from ...models.utils import _copy_tensor
     18 __all__ = ["_AbstractDataset", "_VisionDataset"]
     21 class _AbstractDataset:

File c:\Users\vishw\Desktop\test\doctr\doctr\models\utils\__init__.py:4
      1 from doctr.file_utils import is_tf_available, is_torch_available
      3 if is_tf_available():
----> 4     from .tensorflow import *
      5 elif is_torch_available():
      6     from .pytorch import *  # type: ignore[assignment]

File c:\Users\vishw\Desktop\test\doctr\doctr\models\utils\tensorflow.py:12
      9 from zipfile import ZipFile
     11 import tensorflow as tf
---> 12 import tf2onnx
     13 from tensorflow.keras import Model, layers
     15 from doctr.utils.data import download_from_url

File c:\Python312\Lib\site-packages\tf2onnx\__init__.py:8
      3 """tf2onnx package."""
      5 __all__ = ["utils", "graph_matcher", "graph", "graph_builder",
      6            "tfonnx", "shape_inference", "schemas", "tf_utils", "tf_loader", "convert"]
----> 8 import onnx
      9 from .version import git_version, version as __version__
     10 from . import verbose_logging as logging

File c:\Python312\Lib\site-packages\onnx\__init__.py:77
     73 from typing import IO, Literal, Union
     76 from onnx import serialization
---> 77 from onnx.onnx_cpp2py_export import ONNX_ML
     78 from onnx.external_data_helper import (
     79     load_external_data_for_model,
     80     write_external_data_tensors,
     81     convert_model_to_external_data,
     82 )
     83 from onnx.onnx_pb import (
     84     AttributeProto,
     85     EXPERIMENTAL,
   (...)
    111     Version,
    112 )

ImportError: DLL load failed while importing onnx_cpp2py_export: A dynamic link library (DLL) initialization routine failed."

Details about my environment:

I’m working within a Jupyter notebook (run.ipynb).
I’m testing the docTR model in my local Windows environment.
My project structure looks like this

Im trying to run the docTR model given in the github repo link which i’ve posted and i want someone to help me fix the DLL issue related to ONNX so that i can run docTR models.

also I want to understand:

Whether this error is due to ONNX or the OCR model setup.
How to resolve this DLL loading issue to successfully test docTR models in my VSCode Jupyter notebook.

Answers

- EMEliasMi8859
- September 26, 2024 at 8:21 am
- 0 votes
0
import the class of onxx implementation where onnx runtime is used before any other imports
```
import ArcFace
import ...

ImportError: DLL load failed while importing onnx_cpp2py_export: A dynamic link library (DLL) initialization routine failed.
```
this will prevent complect between dependencies
Login or Signup to reply.

- yeamusic21
- October 20, 2024 at 7:15 pm
- 0 votes
0
This is what worked for me
```
pip install "python-doctr[torch,viz,html,contrib]"  
pip install onnx==1.16.1
```
Overall, the main solution for me was downgrading onnx, which is represented in the line
```
pip install onnx==1.16.1
```
And is the recommended solution for this error from onnx, which I found here: https://github.com/onnx/onnx/issues/6267

Note that I’m running Windows 11, Python 3.12.5
Login or Signup to reply.