when I run test.py,error:cuda out of memory,bacth size :1,num_worker :0, single gpu; about arah-release HOT 7 CLOSED

32g vram about my gpu.i think it should enough

from arah-release.

Memory usage depends on image resolution and the size of the human in the image. Are you running on your own data?

If you really want to render high-res image, you can use low VRAM mode as mentioned in #11

from arah-release.

Memory usage depends on image resolution and the size of the human in the image. Are you running on your own data?

If you really want to render high-res image, you can use low VRAM mode as mentioned in #11

I haved set vram=True,but still cuda out of memory; What makes me suspect is that the error does not occur during training, but during testing

from arah-release.

my train and test dataset of image size is the same

from arah-release.

I find in

from arah-release.

when infer first image ,the code is normal, the second image cuda of out memery

from arah-release.

Sorry for the late reply, I've been busy with courses and haven't got a change to look into the issue.

Not sure how you fixed it, as I updated a potential fix to the memory problem - this resolves some of my use cases, as mentioned here. Maybe try merging the latest code if your problem persists.

from arah-release.