Separate kinect depth edit

Hi everyone I’m trying to figure out how to separately edit kinect components (color, depth and alpha) and how to merge them together afterward but I’m having a bit of an issue.
For example, I extracted the depth map with ‘luminance from depth’ but if I input it in the slot of ‘depth camera/ kinect mesh’ the node doesn’t take it into consideration.
I temporarily solved with a displacement map, but I would like to have some clues about a better workflow.
I also had a bit of problem figuring out what’s the functionality of ‘depthcamera/ kinect color key’ and ‘kinect surface mask’ cause there’s no online documentation about them.
Last thing: is there the possibility of apply negative value to depth mask using color correction nodes? or are we limited to clamped values?

This is the way I approached so far:

  1. Two different ‘depth camera/kinect source’ nodes, first one with default parameters, the second one with ‘output depth’ checked (and post-fx nodes applied).

One ‘colour and depth source’ node with the first ‘depth camera/kinect source’ piped in colour and the second one piped in depth input.

The ‘colour and depth source’ node connected to ‘depth camera/ kinect mesh’.

This way I get notch to recognize the filters applied to depth map but the result is awfully quantised (I guess depth has been treated as a 8bit image?)

I also tried the ‘luminance from depth node’ and checked the ‘depth form luminance’ in ‘depth camera/ kinect mesh’ but there is no way for the modified depth to be read.

  1. This is a workaround, unfortunately with similarly poor results

I tried to connect the edited depth map to a shape 3d plane with a ‘displacement map’ node deformer. Unfortunately, the result is still quantized.

I attach the setup and the comparison between Kinect mesh (left) and displaced plane (right). It looks like the depth map is treated as an 8bit image losing much of its details…
how can I check/control the depth bit of video nodes?

  1. Here you can find a .dfx file with all the setups I tried.
    I added region nodes and labels so everything should be self-explanatory (from left to right)