style-based-gan-pytorch icon indicating copy to clipboard operation
style-based-gan-pytorch copied to clipboard

Published 20 hours ago • verified publisher icon rosinality

CUDA out of memory in generate, even with 1 image

Open jamiejacker opened this issue 6 years ago • 5 comments

Before I start, Huge thank you for the pre trained models; but generating is problematic. Each time I try to run the code this happens:output Surprisingly enough, training works perfectly fine. Using PYCharm on Windows 10 fall creators update with python 3.7; Debugging in console.

jamiejacker avatar Mar 09 '19 13:03 jamiejacker

Maybe you can use torch.no_grad(). I think it will significantly reduce memory consumptions.

rosinality avatar Mar 10 '19 01:03 rosinality

you need 1080 to make it work,

nile649 avatar Mar 13 '19 23:03 nile649

Same issue with GT 1030

dimkoss11 avatar Mar 18 '19 14:03 dimkoss11

Same issue here:

  • tried torch.no_grad(), not working
  • using 1080Ti, not working
  • tried with Pytorch 1.0 and 0.4.1, both not working

Ellenisawake avatar Apr 16 '19 10:04 Ellenisawake

Same issue with TITAN V...... I am wondering that is there anybody have already make it through? I have shrunk down the step size ....it works, but the results are not satisfying.

keishatsai avatar May 29 '19 07:05 keishatsai