0

Looking to convert "FastSeg-large" public model from the OpenModelZoo to use with the "image segmentation" demo. I was able to use the download.py utility to get the original model files downloaded. Now i'm running the following command:

python3 converter.py --name fastseg-large

I'm getting the following error:

Module fastseg_large in C:\Users\david\Desktop\Neural Zoo\open_model_zoo\models\public\fastseg-large;C:\Users\david\Desktop\Neural Zoo\open_model_zoo\tools\downloader\public\fastseg-large/model doesn't exist. Check import path and name

cannot import name 'container_abcs' from 'torch._six' (C:\Users\david\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.8_qbz5n2kfra8p0\LocalCache\local-packages\Python38\site-packages\torch\_six.py)

Looks like i'm getting 2 different errors with the first having trouble finding the path to the original torch model files. I confirmed my path down to the "/model" directory exists. In that directory is another subdirectory "fastseg\model" where the torch model files exist.

The second error i'm not sure about - i double checked that i installed Torch from pip.

Any suggestions?

2 Answers2

0

I’ve validated that running converter.py for FastSeg-large public model is working fine in OpenVINO™ 2021.4.752.

Make sure you run the install_prerequisites_onnx.bat batch file to configure the Model Optimizer for ONNX:

cd “C:\Program Files (x86)\Intel\openvino_2021.4.752\deployment_tools\model_optimizer\install_prerequisites”

then

install_prerequisites_onnx.bat
Rommel_Intel
  • 1,369
  • 1
  • 4
  • 8
  • Thanks for the response! I have run those scripts in the past and i re-ran them just in case again however i'm still getting the same errors when i run the converter utility. I must have something broken or missing in my setup or environment. Do the errors i'm getting point to any other things i may be missing in my setup? –  Nov 30 '21 at 02:35
0

Doing some more web searches i see the error can be due to a newer version of Torch. So i uninstalled Torch and reinstalled version 1.8. That actually failed and i went back to the "pip install torch" without specifying the version. Then i saw torchvision was also mentioned so i "pip install torchvision".

Somewhere along the way i must have cleaned up something related to my torch libraries. Either way that got this to work.