Cosmos-Transfer1-7B-Sample-AV-Data-Example

# **Cosmos-Transfer1-7B-Sample-AV-Data-Example**

[**Cosmos**](https://huggingface.co/collections/nvidia/cosmos-transfer1-67c9d328196453be6e568d3e) | [**Code**](https://github.com/nvidia-cosmos/cosmos-transfer1) | [**Paper**](https://arxiv.org/abs/2503.14492) | [**Paper Website**](https://research.nvidia.com/labs/dir/cosmos-transfer1)

## Dataset Description:
This dataset contains 10 sample data points intended to help users better utilize our Cosmos-Transfer1-7B-Sample-AV model. It includes HD Map annotations and LiDAR data, with no personally identifiable information such as faces or license plates. This dataset is intended for research and development only.

## Dataset Owner(s):
NVIDIA

## Dataset Creation Date:
2025/03/16

## License/Terms of Use:
This dataset is governed by [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/legalcode.en).

## Intended Usage:
This dataset is intended to demonstrate and facilitate understanding and usage of the [Cosmos-Transfer1-7B-Sample-AV](https://huggingface.co/nvidia/Cosmos-Transfer1-7B-Sample-AV) model. It should primarily be used for educational and demonstration purposes.

## Dataset Characterization
- Data Collection Method<br>
* Automatic/Sensors <br>

- Labeling Method<br>
* Automatic/Sensors <br>

## Dataset Format
Modality: HDMap Annotations, LiDAR
Format Classification: Structured data (LiDAR Point Clouds, HDMap Annotations)

## Dataset Quantification
Record Count: 10 sample data points
Feature Count: HDMap annotations, LiDAR data
Measurement of Total Data Storage: 10GB

## Reference(s):
Public Repo Location: https://huggingface.co/nvidia/Cosmos-Transfer1-7B-Sample-AV

## Ethical Considerations:
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).

## Contents
You will see the following folder structure, including several attributes (HDMap / LiDAR data, etc).

### Sensor Data

| Folder | File Format | Description |
|--------|------------|-------------|
| `lidar_raw` | `.tar` | Motion-compensated LiDAR point clouds (10 FPS) |
| `pose` | `.tar` | Camera poses derived from vehicle pose (30 FPS) |
| `ftheta_intrinsic` | `.tar` | Camera intrinsic parameters for each view |

### HDMap Annotations

| Folder | File Format | Description |
|--------|------------|-------------|
| `3d_lanes` | `.tar` | 3D lane boundaries (left and right) |
| `3d_lanelines` | `.tar` | 3D lane centerlines |
| `3d_road_boundaries` | `.tar` | Road boundary annotations |
| `3d_wait_lines` | `.tar` | Waiting lines at intersections |
| `3d_crosswalks` | `.tar` | Crosswalk annotations (polygon format) |
| `3d_road_markings` | `.tar` | Road surface markings (turning arrows, stop lines, etc.) |
| `3d_poles` | `.tar` | Traffic poles (3D points) |
| `3d_traffic_lights` | `.tar` | Traffic lights (3D cuboid format) |
| `3d_traffic_signs` | `.tar` | Traffic signs (3D cuboid format) |

### Dynamic Object Annotations

| Folder | File Format | Description |
|--------|------------|-------------|
| `all_object_info` | `.tar` | 4D object tracking (position, dimensions, movement state) |

## Camera and LiDAR Synchronization
- **Camera Frame Rate:** 30 FPS
- **LiDAR Frame Rate:** 10 FPS
- **Synchronization:** Each LiDAR frame corresponds to **3 consecutive camera frames**.
- **Pose Interpolation:** Camera poses are interpolated at the **starting timestamp** of each LiDAR frame.


## How to Use the Dataset
Please visit https://github.com/nv-tlabs/cosmos-av-sample-toolkits to learn about how to use this dataset example.

# **Cosmos-Transfer1-7B-Sample-AV 样本自动驾驶数据集示例**

[**Cosmos**](https://huggingface.co/collections/nvidia/cosmos-transfer1-67c9d328196453be6e568d3e) | [**代码库**](https://github.com/nvidia-cosmos/cosmos-transfer1) | [**论文**](https://arxiv.org/abs/2503.14492) | [**论文官网**](https://research.nvidia.com/labs/dir/cosmos-transfer1)

## 数据集描述：
本数据集包含10个样本数据点，旨在帮助用户更好地使用我们的Cosmos-Transfer1-7B-Sample-AV模型。数据集包含高清地图（HDMap）标注与激光雷达（LiDAR）数据，未包含人脸、车牌等个人可识别信息。本数据集仅用于研发用途。

## 数据集所有者：
NVIDIA（英伟达）

## 数据集创建日期：
2025/03/16

## 使用许可条款：
本数据集遵循[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/legalcode.en)协议。

## 预期用途：
本数据集用于演示并帮助用户理解与使用[Cosmos-Transfer1-7B-Sample-AV](https://huggingface.co/nvidia/Cosmos-Transfer1-7B-Sample-AV)模型，主要用于教学与演示场景。

## 数据集特征
- 数据采集方式：自动采集/传感器采集
- 标注方式：自动标注/传感器标注

## 数据集格式
模态：高清地图（HDMap）标注、激光雷达（LiDAR）数据
格式分类：结构化数据（激光雷达点云、高清地图标注）

## 数据集量化指标
记录数量：10个样本数据点
特征类型：高清地图标注、激光雷达数据
总数据存储量：10GB

## 参考资源：
公开仓库地址：https://huggingface.co/nvidia/Cosmos-Transfer1-7B-Sample-AV

## 伦理考量：
NVIDIA（英伟达）认为，可信人工智能（AI）是一项共同责任，我们已建立相关政策与实践规范，以支持各类人工智能应用的开发。开发者在按照服务条款下载或使用本模型时，应与内部模型团队协作，确保该模型符合相关行业与应用场景的要求，并应对可能出现的产品误用问题。
请通过[此链接](https://www.nvidia.com/en-us/support/submit-security-vulnerability/)报告安全漏洞或英伟达人工智能相关问题。

## 数据集内容
本数据集包含以下目录结构，包含多种数据属性（高清地图/激光雷达数据等）。

### 传感器数据

| 文件夹名称 | 文件格式 | 描述 |
|--------|------------|-------------|
| `lidar_raw` | `.tar` | 经过运动补偿的激光雷达点云（帧率10 FPS） |
| `pose` | `.tar` | 基于车辆位姿推导的相机位姿（帧率30 FPS） |
| `ftheta_intrinsic` | `.tar` | 各视角的相机内参 |

### 高清地图（HDMap）标注

| 文件夹名称 | 文件格式 | 描述 |
|--------|------------|-------------|
| `3d_lanes` | `.tar` | 3D车道边界（左右两侧） |
| `3d_lanelines` | `.tar` | 3D车道中心线 |
| `3d_road_boundaries` | `.tar` | 道路边界标注 |
| `3d_wait_lines` | `.tar` | 交叉口等待线 |
| `3d_crosswalks` | `.tar` | 人行横道标注（多边形格式） |
| `3d_road_markings` | `.tar` | 路面标线（转向箭头、停止线等） |
| `3d_poles` | `.tar` | 交通杆（3D点格式） |
| `3d_traffic_lights` | `.tar` | 交通信号灯（3D长方体格式） |
| `3d_traffic_signs` | `.tar` | 交通标志（3D长方体格式） |

### 动态目标标注

| 文件夹名称 | 文件格式 | 描述 |
|--------|------------|-------------|
| `all_object_info` | `.tar` | 4D目标跟踪数据（位置、尺寸、运动状态） |

### 相机与激光雷达同步
- **相机帧率：** 30 FPS
- **激光雷达帧率：** 10 FPS
- **同步机制：** 每个激光雷达帧对应**3个连续的相机帧**。
- **位姿插值：** 相机位姿在每个激光雷达帧的**起始时间戳**处进行插值。


## 数据集使用方法：
请访问https://github.com/nv-tlabs/cosmos-av-sample-toolkits 了解本数据集示例的使用方式。