Error: from torch._C import _set_worker_signal_handlers, _update_worker_pids,

[AtefehBaniasadi]

Hi,

First of all, thank you for your work, when I wanted to test your model i faced this error

from torch._C import _set_worker_signal_handlers, _update_worker_pids,
ImportError: cannot import name '_update_worker_pids'

I thought the error may be related to pythorch version so I changed it but the error didn't resolve, can you please guide me on how I can address the problem?

Thanks in advance.

[mazzzystar]

@AtefehBaniasadi
See CSAILVision/semantic-segmentation-pytorch#157 (comment)

[SuJiaKuan]

In README, the author says the model is built in PyTorch 0.4.0, PyTorch 0.4.1. Install PyTorch version 0.4.1 works for me.

If you use Conda, you can install PyTorch 0.4.1 by following command:

conda install pytorch=0.4.1 torchvision==0.2.1 cuda90 -c pytorch

[abbaskali]

if i have pytorch=1.10.0 is it possible to run this code or shall I return to pytorch =0.4.1 ?

[SuJiaKuan]

@abbaskali Please return to PyTorch 0.4.1 because the APIs changes a lot from version 0.X to 1.X

[abbaskali]

@SuJiaKuan thank you a lot sir for your advice, could you help me with another question, I need Pytorch 1.10.0 so do you think is it possible to create another environment on my machine and install PyTorch 0.4.1 to run this code better?
thank you once again

[SuJiaKuan]

@abbaskali Yes, it is possible. Conda is a very nice tool for Python packages and environment management. Under Conda, you can create separate environments for different purposes, for example, one environment for Pytorch 1.10.0 and one for Pytorch 0.4.1.

The tutorial of Conda.

Example commands to create an environment for PyTorch 0.4.1:

conda create --name ridnet python=3.7
conda activate ridnet
conda install pytorch=0.4.1 torchvision==0.2.1 cuda90 -c pytorch

[abbaskali]

@SuJiaKuan thank you a lot for your great explanation, I appreciate it, sir, my cuda is 11.4 is it okay for "conda install pytorch=0.4.1 torchvision==0.2.1 cuda90 -c pytorch"?

[SuJiaKuan]

@abbaskali Yes, it is. The command also installs Cuda 9.0 (cuda90) that is separate from your own cuda (Cuda 11.4), so just feel free and try it.

[abbaskali]

@SuJiaKuan yes sir, i will do it, thank you for your help, God bless you

[abbaskali]

anybody can help me I got this problem:
from torch.utils.data import _utils
ImportError: cannot import name 'utils' from 'torch.utils.data' (C:\ProgramData\Miniconda3\envs\ridnet\lib\site-packages\torch\utils\data_init.py)

[abbaskali]

dear dr.@SuJiaKuan, i am sorry to bother you, this code is available for Ubuntu can I implement it on windows? or should I install the Ubuntu terminal? thank you

[SuJiaKuan]

@abbaskali I'm, a Linux user, so I'm not sure the code works well or not on Windows. Most popular Python / Deep Learning tools (e.g., Conda) support both Linux and Windows, and I guess it is very possible to run the project on Windows. From my perspective, most (Deep Learning) tools are native built on Linux, so it is recommended to learn and develop projects on Linux. :D

[abbaskali]

Dr. @SuJiaKuan thankful for you helping

[abbaskali]

dear dr. @SuJiaKuan
i got this problem could you help me please
ModuleNotFoundError: No module named 'model.edsr'

[SuJiaKuan]

@abbaskali Do you run the code under directory TestCode/code?