Prepare Data for Use with NuRec#

Before you can generate scenes for use with driving or robotics simulation platforms, you must first validate and convert your real-world driving or robotics data to the standardized NCore data format.

About NCore#

NCore, in the context of NuRec, refers to NVIDIA’s standardized data format for autonomous vehicle and robotics datasets. It provides a unified schema for storing sensor data, annotations, and metadata required for neural reconstruction and simulation. The NCore format includes data types that define sensor rig configurations, camera, lidar, and radar setups in the NuRec scenarios, and metadata and session information. Converting data to the standardized NCore format supports consistent, quality reconstruction and the highest quality output through NuRec rendering.

For more information on NCore data types, see NCore Data Schema.

NuRec also requires an auxiliary dataset to reconstruct scenes when you convert and input your own real-world data. This dataset, called NuRec Auxiliary Data, includes the required and optional data types outlined in Generate Auxiliary Data.

Get NCore-standard Data#

Select one of the following paths to get pre-trained scenes, pre-converted data, or raw data from avaialable datasets on HuggingFace to use with NuRec:

  1. Get generated, neurally reconstructed scenes from the NVIDIA NuRec-trained Physical AI dataset and then follow the instructions to Render the Physical AI dataset with NuRec.

  2. Get NCore-converted data from the NVIDIA NCore-converted Physical AI dataset and then Generate Auxiliary Data to generate NuRec-trained scenes.

  3. Get raw sensor data from the NVIDIA Physical AI raw dataset and convert it using the NVIDIA Physical AI Data Converter, then Generate Auxiliary Data, and use that to generate NuRec-trained scenes.

On this page