PhysicalAI-Autonomous-Vehicles-NCore

# PhysicalAI Autonomous Vehicles - NCore

A subset of ~1.1k clips from the [PhysicalAI-Autonomous-Vehicles](https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles) (PAI-AV)
dataset, converted to [NCore](https://github.com/NVIDIA/ncore) format
using the [PAI data converter](https://github.com/NVIDIA/ncore/tree/main/tools/data_converter/pai).
The subset contains clips that expose accurate offline calibration,
egomotion, and cuboid labels.

## Source Dataset

[PhysicalAI-Autonomous-Vehicles](https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles)
contains 306,152 clips (1,700 hours) of multi-sensor driving data
collected across 25 countries. See the source dataset card for full details
on geographic coverage, sensor specifications, and data schemas.

## Contents

Each clip in this repository contains a NCore V4 dataset:

- **Camera sensors**
- **Lidar sensor**
- **Offline calibration** - offline estimated camera and lidar intrinsics,
sensor extrinsics, camera sensor masks, as well as windshield model
parameters if available
- **Offline Egomotion** - offline estimated ego vehicle pose
- **Cuboid labels** - offline estimated 3D bounding boxes

Per-clip file structure:

clips/<clip_uuid>/
├── pai_<clip_uuid>.json
├── pai_<clip_uuid>.ncore4.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_cross_left_120fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_cross_right_120fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_front_tele_30fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_front_wide_120fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_rear_left_70fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_rear_right_70fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_rear_tele_30fov.zarr.itar
└── pai_<clip_uuid>.ncore4-lidar_top_360fov.zarr.itar

## License / Terms of Use

[NVIDIA Autonomous Vehicle Dataset License Agreement](https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles/blob/main/LICENSE.pdf)

## Dataset Owner(s)

NVIDIA Corporation

## Dataset Creation Date

03/06/2026

## Intended Usage

This dataset can be used for **autonomous vehicle related use cases only** which can be both **commercial or non-commercial**
as long as the mentioned license terms are abided by. The size and diversity of this multi-sensor dataset makes it
great for research on end-to-end driving, neural reconstruction, synthetic data generation, scenario mining, and many
other autonomous vehicle applications.

## Dataset Characterization

- Data Collection Method
- Automatic/Sensor

- Labeling Method
- Automatic/Sensor

## References

- [NCore](https://github.com/NVIDIA/ncore) - NCore project
- [PAI data converter](https://github.com/NVIDIA/ncore/tree/main/tools/data_converter/pai) - Tool used to convert from the source format to NCore
- [PhysicalAI-Autonomous-Vehicles](https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles) - Source dataset

# PhysicalAI自动驾驶汽车 - NCore

本数据集为[PhysicalAI-Autonomous-Vehicles](https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles)（简称PAI-AV）数据集的子集，包含约1100段片段，已通过[PAI数据转换工具](https://github.com/NVIDIA/ncore/tree/main/tools/data_converter/pai)转换为[NCore](https://github.com/NVIDIA/ncore)格式。该子集收录的片段均包含精准的离线标定、自运动估计数据以及长方体标注框。

## 源数据集

[PhysicalAI-Autonomous-Vehicles](https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles)数据集共计收录306152段片段，总时长达1700小时，为覆盖全球25个国家的多传感器驾驶数据集合。如需了解地理覆盖范围、传感器规格及数据架构的完整细节，请参阅源数据集卡片。

## 数据集内容

本仓库中的每段片段均为NCore V4数据集：
- **相机传感器**
- **激光雷达传感器**
- **离线标定**：包含离线估计的相机与激光雷达内参、传感器外参、相机传感器掩码，以及可用的挡风玻璃模型参数
- **离线自运动估计**：离线估计的自车位姿数据
- **长方体标注框**：离线估计的3D边界框

单片段文件结构：

clips/<clip_uuid>/
├── pai_<clip_uuid>.json
├── pai_<clip_uuid>.ncore4.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_cross_left_120fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_cross_right_120fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_front_tele_30fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_front_wide_120fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_rear_left_70fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_rear_right_70fov.zarr.itar
├── pai_<clip_uuid>.ncore4-camera_rear_tele_30fov.zarr.itar
└── pai_<clip_uuid>.ncore4-lidar_top_360fov.zarr.itar

## 使用许可与条款

本数据集需遵守[NVIDIA自动驾驶数据集许可协议](https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles/blob/main/LICENSE.pdf)。

## 数据集所有者

NVIDIA公司（NVIDIA Corporation）

## 数据集创建日期

2026年3月6日

## 预期用途

本数据集仅可用于**自动驾驶相关场景**，可用于商业或非商业用途，但需遵守前述许可协议条款。该多传感器数据集的规模与多样性使其非常适用于端到端驾驶、神经重建、合成数据生成、场景挖掘以及诸多其他自动驾驶应用的研究。

## 数据集特征

- 数据采集方式：自动/传感器采集
- 标注方式：自动/传感器标注

## 参考资料

- [NCore](https://github.com/NVIDIA/ncore) - NCore项目
- [PAI数据转换工具](https://github.com/NVIDIA/ncore/tree/main/tools/data_converter/pai) - 用于将源格式转换为NCore格式的工具
- [PhysicalAI-Autonomous-Vehicles](https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles) - 源数据集