Page 1 of 1

set the pipeline locally without using GPU

Posted: Fri Oct 18, 2024 12:09 pm
by jonathancheng
Hello,
I am trying to run the pipeline locally, but since I do not have a desktop with a GPU, I used the TensorFlow CPU (version 2.17.0) instead. However, I encountered a DLL load failure error when importing TensorFlow. Additionally, I tried using Google Colab to run the pipeline, but ran into some version conflicts.

Has anyone successfully set up the pipeline locally using CPU TensorFlow or on Google Colab? Or do I need to consider purchasing a GPU to ensure the program works?

Thank you

Re: set the pipeline locally without using GPU

Posted: Mon Oct 21, 2024 10:23 am
by mpetrucc
Hi Yu-Chen,

If you have already collected data in OpenCap, it would be easier to run the reprocessSession.py code. If you are trying to run the rest of the pipeline without reprocessing the pose estmation, you could skip it by setting it to false here: https://github.com/stanfordnmbl/opencap ... ain.py#L80.

Hope this helps,
Matt

Re: set the pipeline locally without using GPU

Posted: Wed Oct 23, 2024 4:05 pm
by jonathancheng
Hello Matt,
Thank you for your reply.
Currently I have done all the installation process in a desktop with GPU, and the installation looks good.
I am wondering what should be my next step to test whether it can be used locally?(should I first upload some videos in OpenCap or what)

Thank you!

Re: set the pipeline locally without using GPU

Posted: Thu Oct 24, 2024 12:19 pm
by mpetrucc
Hi Yu-Chen,

See the steps in the readme here: https://github.com/stanfordnmbl/opencap ... popencapai

You can either use your own session or reproduce the paper results to test your pipeline.

Hope this helps,
Matt