Getting error when trying to run StyleGAN2 models

[erinbeesley]

Currently trying to play around with this as a little experiment, but whenever I try to run the StyleGAN2 interactive models, I get the following runtime error:

RuntimeError: cuDNN error: CUDNN_STATUS_MAPPING_ERROR

I believe I may not have correctly configured/set up the StyleGAN2 installation from the page, as when I run the command prompt for VS2017, I can't even use Anaconda correctly, but I can get through the other steps fine.

I'm on a RTX 2070 and a Ryzen 5 3600.

[harskish]

Hi, sorry for the late reply.

This seems like a problem with your environment, and not an issue with ganspace per se - I've never seen this error before.
The error you mentioned with anaconda not being usabe in VS2017 is surely part of it. You might have to run conda init in that terminal. Also, make sure that the cuda version you have installed on your system matches the version expected (10.1).

[ghost]

Also see this error with conda installation on windows following the readme.

"...\models\stylegan2\stylegan2-pytorch\model.py", line 273, in forward
    out = F.conv2d(input, weight, padding=self.padding, groups=batch)
RuntimeError: cuDNN error: CUDNN_STATUS_MAPPING_ERROR

Also tried torch.backends.cudnn.enabled = False but it fails in some .cu files.

[ghost]

I tried python visualize.py --model=StyleGAN2 --class=ffhq --use_w --layer=style -b=10_000 and it works nicely. Only interactive.py fails.

[ima9ine]

same here

[negon]

did anybody find a solution for this, I have tried with CUDA 11.x with both NVIDIA 510 and 470 but nothing works, always just CUDNN_STATUS_MAPPING_ERROR.
Card is RTX 3070 with 11GB memory