Talmaj

I personally use pip for all my projects because it is more lightweight, however the nightly version has some features that are not available in the released version (faster mesh_normal_consistency_loss)....

In general it would be nice to have a list of prebuilt wheels available.

I installed the nightly version, which seems to have the C++ implementation. It is definitely much faster. Before I did not get fully utilised GPU on g4dn.x2large, with the nightly...

I am going to prepare a pull request soon.

We've added an example using Pytorch: https://github.com/fraunhoferhhi/DeepCABAC/blob/master/examples/example_pytorch.py You can replace the resnet18 model with VGG16 and adjust the parameters.

Try `docker pull mcr.microsoft.com/azureml/onnxruntime:latest-tensorrt` and re-run. I am not sure if Microsoft is regularly updating these images. Otherwise you can compile and build the image yourself: https://github.com/microsoft/onnxruntime/blob/master/dockerfiles/Dockerfile.tensorrt