These dataset are obtained in realistic department store environments!
(This demo video is the case when the mode is set to both. Please refer to below instructions)
- Complex Indoor environments
- Hundreds of dynamic objects, i.e. humans
- For detailed explanations, please refer to NAVER LABS website
- Data can be downloaded here.
- Unfortunately, as far as I know, HD Map & Localization Dataset request is available only to Korean researchers and organizations... 😭 .
data_path (`/nvlabs/abs_dir` in `launch/lidar_publisher.launch` file)
_____images
|...
_____pointclouds_data
|...
_____camera_parameters.txt
_____groundtruth.hdf5
_____map.pcd
In fact, it's quite complex to parsing data...But don't worry, I did!. You can just load poses and the corresponding lidar data easily as follows:
/***
* How to load the i-th point cloud
*/
pcl::PointCloud<PointType>::Ptr srcCloud(new pcl::PointCloud<PointType>);
*srcCloud = *loader.cloud(i);
/***
* How to load the i-th pose
*/
Eigen::Matrix4f pose = loader.pose(i);Please refer to src/data_publisher.cpp.
$ catkin build naverlabs_api
$ roslaunch naverlabs_api lidar_publisher.launch
Note that you can choose the lidar type among three modes, i.e. lidar0, lidar1, and both.
Also, other parameters are in launch/lidar_publisher.launch.
You can erase traces of dynamic objects by ERASOR.
(The below cyan points denote the traces by humans!)
| Before the application | After the application |
|---|---|
 |
 |