Run
roslaunch rcta_object_pose_detection rcta_object_pose_detection.launch
to start the pose estimation.
For the keypoint base pose estimation, pytorch must be installed.
I run with version 1.7.1.
Pip seems to only offers up to version 1.4 for python2.7, so the keypoint detector node is set to run with python 3.
The pose optimization depends on gtsam.
gtsam does not appear to be able to be built with catkin_make.
Instead, you should use catkin_make_isolated or catkin build.