Hello!
I am trying to build pointcloud2 data in ROS Noetic from the depth images provided in the dataset. In order to do so I am using the equations provided in the dataset documentation:

However I am not able to build a pointcloud that has realistic dimensions. I really don't understand why the value of the pixel has to be divided by 6553.5, what does this scaling factor means? and why is it necessary?

If I take out the scaling factor and use the value of the pixel as depth I get an enormous pointcloud much bigger than the map dimensions. If I leave the scaling factor in the equations then the size of the obtained pointcloud is extremely small, less than a meter.

I have been trying experimentally with other scaling values and the equations that give me a more or less realistic size pointcloud are:

I am implementing a ROS node in python to do this. The values I am using for the camera intrinsic parameters are the ones provided in the documentation:
cx = 157.3245865
cy = 120.0802295
fx = 286.441384
fy = 271.36999
Any help with the correct implementation of these equations will be greatly appreciated!!!!!