Hi - I am trying to recover the depth values from the shift values that can be saved out from STDepthFrame.
I am aware of discussions in http://forums.structure.io/t/raw-data-from-the-sensor-depth-and-ir-frame/2227, as well as uplink's conversion table https://github.com/occipital/uplink/blob/master/headers/core/shift2depth.hpp.
For the table case I find it to underestimate the depths when drawn alongside the raw depth from sensor. In picture below the backprojected depth from sensor is visualized in blue, and converted shift values have black-to-red gradation based on distance from actual depth. Ideally the two should overlap, but clearly it is not the case.
As far as fitting the function - it seems that function proposed in http://forums.structure.io/t/raw-data-from-the-sensor-depth-and-ir-frame/2227 is incorrect as single shift value can map to multiple depth values, when plotting the observations agains one another (See below - x axis is shift values and y axis is depth. There are also multiple visualization of conversion functions I've tried).
So I hypothesize that actual conversion from shift to depth might vary based on the distance from camera and image location. I wanted to ask wheter it would be possible to know how to do the shift to depth conversion correctly?