GAD Editorial · Published · · English study · Chinese summary · Study dated 2026-10-04
A single incorrect connection can change a site's apparent relationship to its neighbourhood. Two roads may cross in an aerial image while remaining separated vertically. A narrow break may conceal a real route. For early site analysis, these small geometric ambiguities can matter more than the visual completeness of the overall map.
Sat2Graph addresses road-graph extraction from satellite imagery. The paper, submitted on 19 July 2020 and identified as ECCV 2020 research, encodes a road network into a tensor that a learned model predicts. The authors also examine stacked roads, including overpasses. This is historical research about extracting network structure; it does not establish a ready-made pedestrian-access map for an architectural project. Sat2Graph paper
Identify the consequential edges
GAD proposes a site-study method that begins with the decision, such as comparing the relationship of two possible entrances to nearby destinations. Draw the candidate routes and identify which connections determine the comparison. A bridge, a crossing or a short access road may be more consequential than many ordinary street segments.
Place those critical connections in a verification list. For each, record what the image appears to show, the network's interpretation and the independent evidence still needed. Preserve the imagery date where available. A road graph extracted correctly from an old image can still be wrong for the present site condition.
Keep geometric connection separate from permission and usability. A visible route does not by itself establish public access, an accessible gradient, a safe crossing or an open entrance. Those are different attributes requiring suitable sources or site investigation. Do not allow the existence of an edge in a graph to answer all of them.
Compare uncertainty as a scenario
Before publishing a catchment or access diagram, test the critical uncertain connections in more than one state. What changes if an apparent link is unavailable? Which destination becomes meaningfully harder to reach? If the entrance ranking reverses, that uncertainty should be prominent in the design discussion.
This is a sensitivity exercise, not a prediction of a closure. Its value is to distinguish an option that depends on one unresolved connection from an option whose apparent advantage survives that uncertainty. Keep the modelled travel assumptions consistent across the scenarios.
Review the network at both the junction and route levels. A plausible-looking intersection may still send a route onto the wrong segment. Conversely, one small drawing error may have little consequence for the particular question. Prioritise correction by the decision it could change, while preserving an honest record of remaining gaps.
Publish the basis of the diagram
A useful presentation includes the chosen entrances, relevant destinations, network source, date and unresolved links. Avoid presenting an inferred road network as a verified walkability assessment.
GAD has not run Sat2Graph or reproduced its benchmark. The proposed workflow takes a lesson from the research representation: connections must be evaluated as connections. For architects, the payoff is a site argument that remains understandable when someone asks the simplest and most consequential question: can people actually get through there?