AM62A7: How can I download edgeai-xvision

Hello Machida-san,

You are correct that this package is not a general package. I believe what you're looking for is part of the edgeai-torchvision github repository. See here: https://github.com/TexasInstruments/edgeai-torchvision/tree/master/references/edgeailite/edgeai_xvision -- I believe these are the files it should be looking for.

The edgeai-mmdetection is often used alongside edgeai-modelmaker. Modelmaker's setup scripts download repositories like edgeai-torchvision and edgeai-mmdetection, and when called, will ensure all paths are setup correctly. References are shared between these repositories, and they should exist within the same parent directory.

If you are using mmdetection as standalone, make sure to have the other repos accessible from the parent directory, and that you've run setup scripts for them. You can model training from the edgeai-mmdetection, but the scripts may require more configuration since they're intended to run from edgeai-modelmaker/run_modelmaker.sh (with a .yaml config file provided as an argument).

I hope this addresses your question and provides enough context for you to continue your development!

Best
-Reese

As Reese explained, edgeai_xvision is located here:

https://github.com/TexasInstruments/edgeai-torchvision/tree/master/references/edgeailite

It is installed when this setup file is run:

https://github.com/TexasInstruments/edgeai-torchvision/blob/master/setup_cpu.sh#L41

Hello,

Thank you for your reply.
Yes, I noticed that "edgeai_xvision" is located in "edgeai-torchvision". However, it seems required python version is difference between "edgeai-torchvision" repo and "edgeai-mmdetection" repo.

If this is correct, I need to configuration two python version and switch version by command.
This will become complex configuration for host machine, so I asked this question.
From your reply, I understand this module should be installed by "edgeai_xvision", so need to create this complex configuration.

Best Regards,

Python 3.10 is what we use now. I think some of the documentation regarding Python version is not updated.  Please try with Python 3.10

Adding onto this: since there are multiple repositories being used in some of these training/model development workflows, the separate respositories will need to be kept up to date. Best practice is to keep the repositories aligned via SDK version tags (e.g. SDK 9.0.0.0). It's possible one of your repos is out of date, such that it's still attempting to use python 3.6 for some dependencies. With SDK 9.0 (released in August 2023), we upgraded to Py3.10 universally. Any dependencies, like an ONNX build, will have a released 3.10 version available (usually setup from a setup.sh script or requirements.txt pip file).

-Reese

Hello,

Thank you for your reply. I will check it.

BR,