Tuning dalponte2016

[ouroukhai]

Hello,

I currently work on a segmentation project. My goal is to tune the window function of the dalponte2016 algorithm. I understand that this can be site and species specific, but does anyone know any bibliography on how to tune the parameters (or estimate them) out of ground truth measurements?
For reference, I have ground measurements on the same plot, from which I can extract distances among some trees (of course with some bias, because I do not have all the trees in the plot).
I also read a very nice example in the book of the package which presented this function:

fws <- function(x) {
  y <- 2.6 * (-(exp(-0.08*(x-2)) - 1)) + 3
  y[x < 2] <- 3
  y[x > 20] <- 5
  return(y)
}

and was wondering if these are estimated parameters or trial-and-error results.

Best!

[Jean-Romain]

There is no windows size parameter in dalponte2016. I guess you are talking about local maximum filter. There is no literature on the subject as far as I know, except the original paper mentioned in the documentation. The function you are mentioning from the book is a pure dummy function used to create an example that is visually explicit.

[ouroukhai]

Thank you very much for the answer and the time. You are correct; I am referring to the local maximum filter window size function. I think I saw in the book some heuristic guidelines for the lower and upper size of the window. Do these come from experience, or is there any bibliography I could find more on, even if they are mostly empirical?

[Jean-Romain]

You have seen some advices to get a code that is working. I'm only providing guidelines to get your code working properly but have no authority to tell people what they should or should do "ecologically speaking". The bounds presented in the book are pure dummy values to take the lead on bugs people may encounter with bad functions (e.g. a function that returns some negative windows size for some Z values)

The lidR package has been design to try, test and explore some stuff from the academic literature. If your code does not work I can help you, but I can't tell you what you should or shouldn't do with a functional code.

[ouroukhai]

Thank you for the explanation.