search

naver-ai / StyleMapGAN

Official pytorch implementation of StyleMapGAN (CVPR 2021)
https://www.youtube.com/watch?v=qCapNyRA_Ng
Other
458 stars 81 forks source link

CUDA out of memory. #4

Closed ykk648 closed 3 years ago

ykk648 commented 3 years ago

图片

How much GPU memory does I need to launch demo.py?

Radium98 commented 3 years ago

Do you encounter the problem“ImportError: /home/f403/.cache/torch_extensions/fused/fused.so: undefined symbol: _ZTIN3c1021AutogradMetaInterfaceE” when you run demo.py? Can you leave your email address I want to ask something about this project. ![Uploading image.png…]()

ykk648 commented 3 years ago

@Radium98 It looks like some kind of 'Environment' issue, follow install.sh, the only different is that I use cuda-11.1 and pytorch-1.8.

Radium98 commented 3 years ago

Have you run the demo.py successfully ?

ykk648 commented 3 years ago

I tried afhq instead of celeba_hq, CUDA out of memory again.

Radium98 commented 3 years ago

It's really frustrating，and I don't know how to process my own data to feed it into the network. Have you solved this problem ?

blandocs commented 3 years ago

@cloisonne GPU consumption mostly depends on input image sizes and _interpolationstep argument. Please use smaller _interpolationstep or input size (pretrained model).

blandocs commented 3 years ago

Do you encounter the problem“ImportError: /home/f403/.cache/torch_extensions/fused/fused.so: undefined symbol: _ZTIN3c1021AutogradMetaInterfaceE” when you run demo.py? Can you leave your email address I want to ask something about this project.

I'm not sure the reason for your problem. Cuda and Pytorch environment may differ from ours.

blandocs commented 3 years ago

It's really frustrating，and I don't know how to process my own data to feed it into the network. Have you solved this problem ?

I just replied. Please check your previous issue :)