Add a scale parameter, to better handle low density pointclouds

[Isaskar]

So I've been trying for the past week to make Roofer work with the Swedish mapping agency Lantmäteriet's open LIDAR data. This is quite low density compared to what Roofer is designed around (only 1-2 points per m2) but it still has enough detail to be useful.

I spent ages trying to find parameter values that worked but I only got garbage results. That is until I finally had the idea of scaling my input data down, to simulate a higher density. Amazingly, this actually worked, and for the first time I got usable results. My testing indicates that a scale factor of 0.5-0.6 is optimal for the Lantmäteriet data set.

Now, I am of course able to do this by scaling my inputs down before use and then scaling the output inversely, but it would be much easier if Roofer could do this for me. How this is best implemented I don't know, but at least to me a scale parameter similar to what I'm already doing seems like a good idea.

For testing purposes, I will include some files here.

• Sample point cloud and footprints from Lantmäteriet (Note: unless you preprocess you must use --bld-class 1): Lantmäteriet data.zip
• Sample output without scaling 321528_6393936.city.json
• Sample output with input scaled down by 0.55 (With the output scaled back up afterwards): 176840_3516664.city.json

Here are some images to illustrate just how much of a difference this makes:

No scaling of input

Input scaled by a factor of 0.55

[Ylannl]

Thank you, that's quite interesting! I will have a look.

[Ylannl]

Version 1.1 will have a unit-scale parameter that can be uses for this purpose.