GAD Editorial · Published · · English study · Chinese summary · Study dated 2026-10-05
A complete neighbouring building makes a site model easier to use. It can also conceal how little of that neighbour was observed. An airborne scan may describe the roof more clearly than the facade, while trees and viewing angles leave other regions sparse. Once those points become a continuous mesh, the missing evidence becomes less obvious.
Point2Building addresses reconstruction from airborne LiDAR by generating sequences of mesh vertices and faces. Its preprint was first submitted on 4 March 2024 and revised on 29 July 2024. The authors identify uneven point density and incomplete facade coverage as central difficulties, and report evaluation using data from Zurich, Berlin and Tallinn. These are research findings, not a guarantee for a particular site's context model. Paper and version history
Give context an evidence scale
GAD proposes a context-model review centred on the design decision. A neighbouring volume used only to orient a concept diagram needs a different evidence threshold from an obstruction that strongly affects a daylight comparison. Classify the intended uses before deciding which reconstructed features need closer investigation.
Keep the source point cloud available with the mesh. In a comparison view, identify surfaces strongly supported by observations and regions whose geometry is mostly inferred. The aim is not to reject every completion. It is to stop the convenience of a closed envelope from silently upgrading its authority.
Mark the capture date, coordinate reference and processing revision. A correctly reconstructed building from an earlier survey can still be inappropriate evidence for a recently changed site. Separate this question of currency from the question of geometric fit.
Inspect the neighbours that change the answer
For a hypothetical courtyard study, first identify the adjacent height or setback that determines whether one option looks preferable. Review that feature directly against the available survey evidence. A citywide average reconstruction score says little about the one edge governing this comparison.
Run a bounded sensitivity study when a consequential dimension remains uncertain. Compare plausible alternative context assumptions supported by the evidence, and record whether the preferred design changes. If it does, the team has located a survey priority rather than established a winning scheme.
Also distinguish geometric omissions from project information the method does not attempt to supply. A reconstructed envelope does not establish a legal boundary, facade opening schedule, material assembly or internal floor level. Those facts should come from appropriate records and measurements.
Preserve useful incompleteness
A review model can contain a deliberately simplified or unresolved neighbour. That may be more responsible than supplying unsupported detail. Use clear object-level notes so later exports and presentation views do not lose the qualification.
GAD has not reproduced Point2Building's experiments. A practice considering it would need to test the relevant building types and capture conditions. The transferable lesson is a disciplined relationship between geometry and evidence: complete the model only as far as the task requires, and keep the important uncertainties available to the next designer.