Some options get deprecated as we update and find optimal defaults.
You can try relaxing --pc-filter, keeping in mind it represents standard deviations from the mean and is as such not linear. Something like --pc-filter 5 or 10 might make a bit of a difference.
Happy New Year!
You may be able to process only the images that contain the conflicting tree or area of interest. With no less than 5 images with something in common, you could process to adjust parameters.
Here are some results from using ‘PC filter’ 5 & 10. No real change so I guess it’s a question of obtaining many more images from the orbitals to give more detail in the sides and therefore to increase the density of those points in the point cloud. There is no shortage of detail in the model but no height information has been derived to position the relevant points correctly.