This is potentially useful to others trying to quickly get set up with this repository and doesn't necessarily need to be merged with master unless the authors see it fit,
I've created a quick setup Dockerfile that uses CUDA 10.0, PyTorch 1.0.0, CuDNN 7.0 and installs all the necessary dependencies to at least run the evaluation script (training untested).
The evaluation script would need to be run with Python3.5 if you use this Dockerfile.
This is potentially useful to others trying to quickly get set up with this repository and doesn't necessarily need to be merged with master unless the authors see it fit,
I've created a quick setup Dockerfile that uses CUDA 10.0, PyTorch 1.0.0, CuDNN 7.0 and installs all the necessary dependencies to at least run the evaluation script (training untested).
The evaluation script would need to be run with Python3.5 if you use this Dockerfile.
I've tested it on a DGX station that's using the NVIDIA 418.67 driver and it works.
Hopefully this is useful to others.
Best regards, Shreyas