graviti logo产品公开数据集关于我们
登录
5K
3
76
DeepRoute Open Dataset
概要
讨论
代码
活动
V3.0.0
8cdf886·
Jun 30, 2021 4:42 AM
·3Commits
add cover

Overview

DeepRoute Open Dataset is a dataset for the training of level 4 self-driving algorithms. It contains real-world testing scenarios on complex urban roads in Shenzhen, Wuhan, and Hangzhou. We share data to enhance the connection between the research community and the industry, thus accelerating the development and application of self-driving technology.

Data Collection

Data are collected from public urban roads in three different cities in China, which are Shenzhen, Wuhan, and Hangzhou. Currently, the data are LiDAR point clouds collected by three LiDARs which are set on the top and sides of the testing vehicle.

Data Annotation

The data released were randomly selected from 10,000 frames of road testing data from three cities. Using the joint annotation with images, the obstacles were annotated by precise 3D bounding boxes. We annotated eight types of obstacles. In each type of the obstacles, we have information including the object type, location of the bounding box's center, dimension of the bounding box, and rotation angle of object's heading.

Please note that when one obstacle is partially observed, the markers are asked to try to restore the full size of obstacles through imagination or sequence and image information, so the real size of the objects can be reflected.

Data Format

There are N*4 floating-point data representing N points in each frame of point cloud data. Each point has the x, y, z (Cartesian coordinates), and normalized intensity data. Each frame of point cloud corresponds to an annotation JSON file. The annotation JSON file contains a list of obstacles, such as:

    "objects": [
        {
            "type": "CYCLIST",
            "bounding_box": {
                "width": 1.100000023841858,
                "length": 2,
                "height": 1.7999999523162842
            },
            "position": {
                "y": -22.889999389648438,
                "x": 7.059999942779541,
                "z": -0.75
            },
            "heading": 5.98647928237915,
            "id": 10
        },
        ...
    ]

Instruction

1. Load the point cloud data

def load_pointcloud(pc_file):
    pc = numpy.fromfile(pc_file, dtype=np.float32)
    return pc.reshape(-1, 4)

2. Load the Groundtruth data

def load_groundtruth(gt_file):
 with open(label_file, "r") as hd:
     objects = json.load(hd)
 return objects["objects"]

Citation

@misc{
Author  = {DeepRoute.ai},
Title = {The DeepRoute Open Dataset for Autonomous Driving},
Year = {2021},
}
数据预览
查看数据
🎉感谢Graviti Open Datasets的贡献
数据集信息
应用场景Autonomous Driving
标注类型ClassificationBox3D
任务类型暂无
LicenseCustom
更新时间2021-03-24 23:07:37
数据概要
数据格式Point Cloud
数据数量10K
已标注数量11999
文件大小13GB
版权归属方
DEEPROUTE.Ai
标注方
DEEPROUTE.Ai
了解更多和支持
立即开始构建AI
免费开始联系我们