myron
myron
the version with nn.functional.one_hot was much slower if someone can retest it (time it for 3D images) , we could consider the change but I see now reason to change...
>If you are OK to save the data in uint8, you can put a CastToType transform That's a good suggestion, it should be smaller in size then, but I guess...
ok, closing for now
this looks good, but shouldn't we support all modes?
same issue on Nvidia 3090 Ubuntu 22.04.1 LTS, Driver Version: 510.85.02 CUDA Version: 11.6
This is on NVIDIA V100 16gb x 8 ngc instance, using NVIDIA pytorch contrainer (either nvidian/pytorch:22.09-py3 or nvidia/pytorch:22.0-py3) and latest MONAI 1.0.0 (via pip install monai) ``` Printing MONAI config......
yeah, you're right 22.08 pytorch container is working fine, which includes ``` CUDA version: 11.7 cuDNN version: 8500 Pytorch version: 1.13.0a0+d321be6 ``` So it's related to newer cudnn or newer...
Thanks @Nic-Ma . In my analysis, monai=0.9.0 works fine in the newest pytorch container, so it's something specific for monai=1.0.0 it seems that this header import alone causes it from...
I get the same error if I import `from monai.config.type_definitions import NdarrayTensor` instead of `import cv2` and that import doesn't import cv2, so it seems there are several ways to...
@Nic-Ma thanks for the reply. I see.. We should reconsider this logic. If someone wants to import only a small component, why do we need to import Everything. This seems...