ImportError: cannot import name 'VOCABS' from partially initialized module 'doctr.datasets'

[yerin723]

I'm trying to make a model out of the training data I labeled.
However, the following error keeps occurring.
ImportError: cannot import name 'VOCABS' from partially initialized module 'doctr.datasets' (most likely due to a circular import) (C:\Users\PIAI\Desktop\ocrproject\doctr\doctr\datasets_init_.py)
Output is truncated. View as a scrollable element or open in a text editor. Adjust cell output settings...

python "C:\Users\PIAI\Desktop\ocrproject\doctr\references\detection\train_pytorch.py" "C:\Users\PIAI\Desktop\test" "C:\Users\PIAI\Desktop\val" db_resnet50 -- epochs 5 --device 0

An error occurs when executing the above command.
Please tell me the solution!

[dvivarelli]

Same problem here

[hanshupe007]

I have the same problem here, using Windows and Python 3.9.

According to the traceback below, the init.py in doctr.datasets (doctr.datasets import VOCABS) itself loads indirectly VOCABS from mobilenet\pytorch.py, which leads to the circular import error.

Traceback (most recent call last):
  File "C:\xx\doctr\references\recognition\train_pytorch.py", line 33, in <module>
    from doctr.datasets import VOCABS
  File "C:\xx\doctr\doctr\datasets\__init__.py", line 3, in <module>
    from .generator import *
  File "C:\xx\doctr\doctr\datasets\generator\__init__.py", line 6, in <module>
    from .pytorch import *  # type: ignore[assignment]
  File "C:\xx\doctr\doctr\datasets\generator\pytorch.py", line 8, in <module>
    from .base import _CharacterGenerator, _WordGenerator
  File "C:\xx\doctr\doctr\datasets\generator\base.py", line 14, in <module>
    from ..datasets import AbstractDataset
  File "C:\xx\doctr\doctr\datasets\datasets\__init__.py", line 6, in <module>
    from .pytorch import *  # type: ignore[assignment]
  File "C:\xx\doctr\doctr\datasets\datasets\pytorch.py", line 15, in <module>
    from .base import _AbstractDataset, _VisionDataset
  File "C:\xx\doctr\doctr\datasets\datasets\base.py", line 16, in <module>
    from ...models.utils import _copy_tensor
  File "C:\xx\doctr\doctr\models\__init__.py", line 1, in <module>
    from .classification import *
  File "C:\xx\doctr\doctr\models\classification\__init__.py", line 1, in <module>
    from .mobilenet import *
  File "C:\xx\doctr\doctr\models\classification\mobilenet\__init__.py", line 6, in <module>
    from .pytorch import *
  File "C:\xx\doctr\doctr\models\classification\mobilenet\pytorch.py", line 13, in <module>
    from doctr.datasets import VOCABS
ImportError: cannot import name 'VOCABS' from partially initialized module 'doctr.datasets' (most likely due to a circular import) (C:\xx\doctr\doctr\datasets\__init__.py)

[felixdittrich92]

Hi @dvivarelli @hanshupe007 👋,

I have had the chance to test it on a windows machine.
Unfortunately i was not able to reproduce it...

Workflow:

• Setup miniconda -> create env
• Clone the repo
• Install pytorch via conda (optional) - on windows it's easier with conda because of precompiled cuda/cudnn - GPU support
• open repo in VSCode -> pip3 install -e ."[torch]" -> pip3 install wandb
• move into recognition folder
• python train_pytorch.py crnn_mobilenet_v3_small

[hanshupe007]

I didn't use conda, but WinPython, did you also use Python 3.9? Looking at the traceback the error makes sense to me, as the __init__py is triggered by the datasets import, and itself imports VOCABS, not sure why it doesn't throw the circular import error on other systems.

[hanshupe007]

While I couldn't resolve the error on my Windows/Python 3.9 environment, it worked on Linux/3.10.

[felixdittrich92]

Hi @hanshupe007 👋,

Thanks for the feedback.
Good to hear that it works now.
But still strange - because as mentioned i haven't had any trouble also on Win :/
maybe a issue with WinPython ..
In general i can only suggest conda (miniconda) - makes it much easier to handle venvs