search

Flode-Labs / vid2densepose

Convert your videos to densepose and use it on MagicAnimate
MIT License
977 stars 126 forks source link

Some conflict and error during generations #8

Closed charliebrown777 closed 9 months ago

charliebrown777 commented 10 months ago

微信图片_20231205232900 微信图片_20231205232938

I got some conflict error reports when installing the repository, but I can get it to run successfully. Some errors show up in the 2nd image and fail to generate the result. Any solutions to that? Thanks!

tadeodonegana commented 10 months ago

Hi @charliebrown777,

The first error seems to be a dependencies issue. Please create a virtual environment to install the required dependencies and avoid dependency conflicts, it is a good practice and it will also help us try to determine a solution for your issue.

For the second error i would suggest installing (or reinstalling) Pytorch with CUDA, here it's an easy-to-follow tutorial that i believe will help.

Please let us know if this works.

chen-rn commented 10 months ago

I am running into the second error that you're running into. @tadeodonegana Is it possible to run this without CUDA purely on the CPU?

tadeodonegana commented 10 months ago

I am running into the second error that you're running into. @tadeodonegana Is it possible to run this without CUDA purely on the CPU?

Hi @chen-rn ,

After setting the model weights in line 22 add the following line:

cfg.MODEL.DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

There's a PR that currently proposes this as a solution, you can see it here #11. We are currently waiting to test the change properly and evaluate merging it.

Please let me know if this works.

tadeodonegana commented 9 months ago

I am running into the second error that you're running into. @tadeodonegana Is it possible to run this without CUDA purely on the CPU?

Hi @chen-rn ,

After setting the model weights in line 22 add the following line:

cfg.MODEL.DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

There's a PR that currently proposes this as a solution, you can see it here #11. We are currently waiting to test the change properly and evaluate merging it.

Please let me know if this works.

This PR was just merged, and it should solve the issue for the app.py file (Gradio version).

tadeodonegana commented 9 months ago

CPU support was also added in #16 and the error about Torch not compiled with CUDA enabled should be now solved.