6d pose estimation --- make your own LINEMOD dataset process record (labelfusion)

After testing many methods, I finally came up with an effective method, see the following link:

effective method

--------------Dividing line---------------

0. Environment configuration

Just configure it according to the method in the connection .

problems encountered

Question 1

When installing ros, pay attention to the ubuntu operating system version, librealsense2 and driver and realsense-ros package version, there may be a situation where the realsense-ros package version fails to compile, you need to change the 44th of CMakeLists.txt under realsense-ros/realsense2_camera/ Be the version number of librealsense2 installed by yourself, mine is 2.45.0.

Question 2

Reference link: https://blog.csdn.net/weixin_41837701/article/details/122668376?spm=1001.2014.3001.5502

1. Start the realsense camera

roslaunch realsense2_camera rs_rgbd.launch

2. Start the rgbd_ros_to_lcm node

roslaunch rgbd_ros_to_lcm lcm_republisher.launch

3. record

Create a log folder /data/logs in any directory, such as:

Run the following command

lcm-logger

A file will be generated after the run

 lcmlog.2020-*

cut out

Open the terminal in this directory, enter the folder where the run_trim script is located and the file, such as:

/home/lab502/LabelFusion/scripts/bin/run_trim

After that, the terminal will prompt you to enter the start time and end time:
the strange thing is that there is no trimmedlog.lcmlog file in my folder, I don’t know what’s going on.

Replenish

The cutting process can actually be carried out in the docker officially provided by labelfusion without error reporting. But when running the run_prep command, there will be an error in Q2 in this link . The reason is roughly like this. My computer graphics card is an NVIDIA TITAN RTX graphics card. According to the table below, my graphics card has a computing power of 7.5. In the official docker The installation is cuda8.*, and the maximum supported computing power is 6.1, so there will be problems.

For this problem, my solution is: don't do it...
try to use the ObjectDatasetTools method , and it works.