Giters

sagiebenaim / DistanceGAN

Pytorch implementation of "One-Sided Unsupervised Domain Mapping" NIPS 2017

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

CycleGan architecture code not CUDA enabled

harsmac opened this issue · comments

commented

I am trying to run my own data using the train.py file which uses the cyclegan architecture. I am trying to run it on a GPU and all libraries are of the same version as you have mentioned. Despite which, I get the following error:
AssertionError: Torch not compiled with CUDA enabled

The command I am running is:
python train.py --dataroot my_data --name exp_1 --model distance_gan --loadSize 128 --fineSize 128 --batchSize 6 --norm instance --A_to_B --nThreads 12 --gpu_ids 0,2

But when I run it with gpu_ids as -1:
python train.py --dataroot my_data --name exp_1 --model distance_gan --loadSize 128 --fineSize 128 --batchSize 6 --norm instance --A_to_B --nThreads 12 --gpu_ids -1
It starts running.

Seems the code was not made for running on GPU??

commented

Hi,

This error seems to be from your installation of pytorch. It basically means that the binaries of pytorch you have do not support GPU. See here pytorch/pytorch#3717 for example.