Dynamic Objects

[MAli7319]

Hi. Thanks for sharing this work as open source. I have just examined the paper and tried to find out that you have tested this project in the outside as well. Are your tests also includes dynamic objects and if yes, how was the results? I m asking because I could not see the related part in the paper or I could overlooked.

[jianhengLiu]

There is one dynamic scene, the FAST-LIVO2 Drive Scene. The dynamic object removal result is shown in the supplementary part of the paper, but it is not the contribution we would like to highlight. As we haven't fully tested the system in the dynamic scenes. If you have some results on these scenes, we would also like to see the results, and we could find some solutions together.

[MAli7319]

Thanks for the answer. At first, we have to consider trying out this project with our self collected data. How can we prepare our data (format, directory sequence etc.) to extract a map afterwards? We now have point clouds, RGB images from stereo, depth, imu and gps for some scenes.

[jianhengLiu]

If you already have your localization results, our system needs RGB images, RGB image poses, point clouds, and point cloud poses. Then you can arrange the data as the 'For Parsered Data' part shown in https://github.com/hku-mars/M2Mapping?tab=readme-ov-file#32-fast-livo2-datasets. Then run as for example (https://github.com/hku-mars/M2Mapping?tab=readme-ov-file#4-run):

# FAST-LIVO2 (Parsered ROS bag format)
    ./devel/lib/neural_mapping/neural_mapping_node train src/M2Mapping/config/fast_livo/campus.yaml src/M2Mapping/data/FAST_LIVO2_RIM_Datasets/campus/color_poses.txt
    # If ROS is installed, you can also run the following command:
    # rosrun neural_mapping neural_mapping_node train src/M2Mapping/config/fast_livo/campus.yaml src/M2Mapping/data/FAST_LIVO2_RIM_Datasets/campus/color_poses.txt

[MAli7319]

For every RGB frame and depth info, there is a 4x4 matrix to indicating the position as far as I understand. We do not have these data yet. Are these positions of camera and lidar with respect to the world or some other frame and how can we get it?

[jianhengLiu]

Please refer to https://github.com/hku-mars/M2Mapping?tab=readme-ov-file#33-custom-fast-livo2-datasets

[MAli7319]

The repo you referred can provide sample scenarios recorded in rosbag with your equipment. That's why ros topics are quite different from ours. So it will not fit if we do not download the one of these custom rosbags. May you propose a different way to solve it?

[jianhengLiu]

You can arrange the data format as shown in the 'Parsered Data' section at https://github.com/hku-mars/M2Mapping?tab=readme-ov-file#33-custom-fast-livo2-datasets.
The poses are the sensor poses with respect to the world coordinate.
For pose estimation, we are now only offering the FASTLIVO2 https://github.com/hku-mars/M2Mapping?tab=readme-ov-file#33-custom-fast-livo2-datasets.
Other pose estimation methods are also supported as the M2Mapping considers pose estimation as the prior input, just arrange the data as the given format.