kolinguo/openchisel repository overview

Container for dense 3D mesh reconstruction from RGB-D observations and the estimated camera pose trajectory. Uses OpenChisel⁠ to reconstruct the scene.
This docker image has Ubuntu 18.04, CUDA 11.3.1, ROS Melodic, ros-melodic-pcl-ros⁠.

⁠Build & compile codebase

1. ./docker_setup.sh openchisel to pull the docker image and build the container.
2. cd scene_recon/openchisel && ./build_ros.sh to compile the packages.

⁠Running

Run ./scene_recon/openchisel/run_openchisel.sh <bag-path> to start the reconstruction from the rosbag (with the estimated camera pose recorded in /tf topic from map frame to imu frame) at <bag-path>. The generated PLY mesh is saved as /maya-slam/scene_recon/output_plys/{bag-name}_openchisel.ply

The script will launch the launch file launch_realsense_maplab.launch⁠. The TSDF <voxel_resolution_m> parameter is specified in meters and can be tuned given the available system RAM resources. However, OpenChisel seems to not function properly when <voxel_resolution_m> is below 0.04 meter (i.e., 4 cm), presumably due to suboptimal implementation.