Very often, I find that parts in the cloud dataset are improperly colored, mostly trees, due to projection of traffic signals, small buildings or else, hence, I need to reconstuct the color.
I'm able to get somewhat decent ouputs, here is a rough example, I recolored areas that did not need recolor but I did so to get a quick example.
what I do is a simple recolor, merge with PCV and Intensity. Finish up with colors levels if needed.
I would like to go bit deeper, while vegetation almost gets a pass with this approach, trunks and branches or mand made objects feel like need something "moar"...
For extracting trunks and branches, I compute geometric features, eigenvalue2 , yet depending on thickness of branches 1 may work better, surface variation or combination of theese + final filter by density outputs a rather leaf free tree structure.
To recolor theese, the above approach leaves to desire. So what I attemped was to create a color scale with multiple color entries, all in the brownish region and colorize with that the intensity SF, and while not epic it started to look more "realistic". Then an idea came to my mind...
I think it is not possible right now, but.. would it be possible tooo create a color scale, based on a chunk of data that is properly colored and, create such color scale based on the intensity values? lets say, the points with 0 Intensity value have this color (wethere is the first color to be found or the average of all the colors in that intensity value) Intensity 1, this other color etc untill the colro scale is filled with as many step as requested or as many Intensity values are present.
So, fasically "Color field to Color scale" or "Scalar field to Colro scale", with an option to define step count or "0" for as much as needed where color position is determined by the Intensity value, or may be dropdown for SF .
Then, on an object of similar nature, lets say the original data was a guardrail it would be quite likely that it would look quite similar to a properly colored object.
What do you think?!