2026/05/25

Results since previous meeting

• Worked on the final MSc thesis report, mostly on its paper part
• Made manual annotations to create a ground-truth dataset of roofprints (no footprints for now)
• Implemented some scores to evaluate the method based on this ground-truth dataset

Material for discussion

Criteria to evaluate the polygons

I have currently implemented two criteria:

• Intersection over union (IoU)
• Chamfer distance

Since the polygons are sometimes adjacent and the shared edges can be impossible to find using only geometrical criteria, I think that it could be beneficial to also compute these same criteria after dissolving the polygons which are connected in the ground-truth, in order to evaluate the method on the level of building blocks instead of parcels.

Visualization of the results

Visualisation

Discussion

Work until next meeting

• Prepare the presentation for the monthly meeting on May 18th
• Look at the two ideas to evaluate the algorithm:
  • Ground-truth dataset of roofprints and footprints
  • Synthetic dataset of LoD 2.3 buildings
• Integrate roofer into the process to filter the points for the footprints
• Improve the footprint generation algorithm to handle the whole building at once