How can I enable Difix (Fixer) for training/testing?

Omniverse Core Platform NuRec
1

Hi, thank you for making the Nurec framework publicly available. I am currently working on 3D reconstruction using the nuPlan dataset, which I have successfully converted to the Ncore format.

I am encountering a blocker when attempting to use the Difix option. While scene reconstruction works correctly with difix.training.enabled=false, enabling the option (difix.training.enabled=true) triggers the following authentication/access error:

[ERROR] Failed to download model: 402 Client Erorr: Payment Required for url: ``https://api.ngc.nvidia.com/v2/org/nvidia/team/nre/models/nurec-fixer/versions/cosmos_3dgut/files/cosmos_3dgut.pt

I have also tried downloading the model via ngc registry following Use the NuRec Fixer Model — NVIDIA Omniverse NuRec , but it also failed. Rendering attempt through Docker cmd or gRPC API with Difix enabled also failed with the same error.

Any guidance on the proper authentication flow for Difix would be greatly appreciated.

Thanks

3

@sungheon.park ,

It seems that you need to have EA access,
In next release, it doesn’t require to have EA access.

In the meantime, You might want to download from huggingface directly.

nvidia/difix · Hugging Face

4

We also have it here
Fixer - Improving 3D Reconstructions with Single-Step Diffusion Models from GitHub - nv-tlabs/Fixer: Official Repository for the NVIDIA Fixer model · GitHub