2026/05/11

Author

Alexandre Bry

Published

May 11, 2026

Results since previous meeting

  • Mainly worked on the presentation and the report
  • Started experimenting on using roofer with my roofprints

Material for discussion

Presentations of my work

We previously discussed about making presentations for other people at IGN, one that would be technical and one that would be more high-level. We were thinking of having the non-technical one on June 22nd and the other one before the A3 presentation for TU Delft which is on June 12th.

Also I was contacted by someone internally at IGN to discuss how I use the LiDAR HD dataset in my work.

It would be great if we could decide on dates for these presentations, keeping in mind that the deadlines for May and June are as follows:

  • May 6th: a short presentation for the LASTIG internal seminar (not too much details in it)
  • May 18th: a monthly presentation, but since new people were invited, should we repurpose it to a general presentation?
  • June 5th: the MSc thesis report (introduction + preliminary materials + paper + discussion)
  • June 12th: the A3 presentation for TU Delft (technical one)
  • June 26th: the A4 presentation for TU Delft (non-technical one)

Discussion

  • It seems better to do roofprint -> roof -> footprint instead of roofprint -> footprint -> roof, because it is easier to identify the points to use in the footprint once we have the roof. Therefore, I should look into using roofer with my footprints to get roofs that I can use to filter the points I want to use for the footprints. This could also help to actually produce 3D buildings, and I could then use my footprints to produce LoD 2.3 buildings!
  • It could be useful to have a look at the arguments of roofer for the complexity of the roofs in order to get roofs that are better to filter out the points for the footprints.
  • We should have a discussion during the monthly meeting on May 18th to decide whether a potential technical presentation in June would be useful, and if so, when it should be.
  • We see two main ways to evaluate the algorithm for the footprints:
    • Create a ground-truth dataset of roofprints (and potentially footprints) and compare these to the results of the algorithm, using metrics such as IoU or Chamfer distance
    • Generate a synthetic dataset of LoD 2.3 buildings from an internal roof dataset using HELIOS++, and then perform the whole process on this dataset to see if we can get good roofprints and footprints. Roof overhangs could be simulated by moving the roofprints towards the inside of the building by a random amount.

Work until next meeting

Back to top