Buildings with workable pointclouds get assigned pointcloud_unusable

[NomadicDaggy]

Is there a way that I can find out why Roofer assigned rf_pointcloud_unusable=true to a building?

Loglevel debug AFAIK doesn't show anything relevant to this. The cropped lidar data for the building looks fine.

Here is an example with a building that clearly has a visible roof without any large holes and good point density and yet Roofer doesn't generate a model for it:

If tilesize is the default 1km, about 10% of the buildings turn out like this and they are equally distributed throughout the tile, not just on edges. If I increase the tilesize, the problem seems to diminish but for a larger dataset even 3km tilesize just stops mid-processing without an error and doesn't generate any cityjson.

Any way I can configure and inspect how pointcloud usability gets assigned?

Still running this version though: roofer a65e086 (main, dirty, 2025-05-30 12:26:47 +0200)

[LeandreSassi]

Same problem here. When I used an older version it worked fine while having ceil-point-density set to 50. Now running with the same toml, lots of polygons only have a lod0.
Putting ceil-point-density down to 1 gets more polygons processed, but still some are skipped and quality is impacted.

I'd like to know if this is normal? Or do I need to fallback to an older version?
In that case, could someone send me the older SHA so I can pull the older docker image?

Thank you!

[Ylannl]

Thanks for reporting this. This attribute is set if the point density of a building is twice the standard deviation less than the mean point density of the tile. The rationale behind this is to detect buildings with footprints are largely occluded/are missing a lot of points (eg. for underground structures, which appear in Dutch data). One assumption here is that the point density is homogeneous over the tile, which is not always true as you found out and this is definitely an unwanted side effect.

I will think of a way to disable this behaviour.

@LeandreSassi unfortunately there is no older version that will 'fix' this.

[LeandreSassi]

Thank you for your response Ylannl, this is very interesting. I'm performing some hyperparameter optimisation to find what variables can prevent this from happening too much. I will try to share any results if that could be of any help.

[Ylannl]

This issue seems to be fixed now with the flag '--no-clear-insufficient', see #133. Please try it with the latest release.