Not working with CPU getting error "OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root." about gpen

[rubo77]

set the CUDA_HOME manually:
https://stackoverflow.com/questions/46064433/cuda-home-path-for-tensorflow

but it seems, I have to install python3-pycuda, which is about 4GB large.

is this needed?

[yangxy]

I have never tried to run GPEN with CPU. I will have a try after my annual leave.

[rubo77]

How so you decide if you use CPU or GPU?
(I have an Nvidia and an internal Intel graphics)

[yangxy]

I am afraid GPEN does not work with CPU in the current form. Please refer to https://github.com/rosinality/stylegan2-pytorch.

[rubo77]

I have an NVIDIA GPU in my laptop, so How can I tell the script to use it? And what are the needed debian packages I have to install to get it running?

You should add some more information to the Readme about the needed packages

[yangxy]

I have an NVIDIA GPU in my laptop, so How can I tell the script to use it? And what are the needed debian packages I have to install to get it running?

You should add some more information to the Readme about the needed packages

Please check the README.md carefully and make sure you have installed python==3.7.4, pytorch==1.7.0, cuda==10.2.89, NVIDIA driver==460.73.01, gcc==7.5.0. You should also install packages listed in the requirements.txt.

[rubo77]

I have gcc 8 installed, are those the minimum requirements >= or must they be exactly that package == ? what is the package-name, you have to install for cuda? there are 20 different packages containing "cuda" is the needed package nvidia-cuda-toolkit? And I only have a package python3-torch

[rubo77]

I see! You need pip to install the required packages:

pip install -r requirements.txt

But I found out, that you need A LOT OF DISK-SPACE for this installation:

ninja:
1.4 MB
numpy:
120 MB
opencv-python:
724 MB
pillow:
137 MB
scikit-image:
7.4 MB
torch:
4772 MB
torchvision:
153 MB

which sums up to about 6GB!!

Would it be possible to minimize this somehow?