UnderOneFacade: Worldwide Facade Semantic Segmentation Benchmark Dataset

Globally consistent semantic digital twins require centimeter-accurate and geographically transferable 3D facade segmentation. How-ever, progress in facade parsing is limited by the lack of large-scale, stan-dardized benchmarks for evaluating cross-domain and cross-country gen-eralization. Existing datasets are geographically narrow, sensor-specific,semantically inconsistent, or insufficiently precise. We introduce Un-derOneFacade, the largest cross-continental 3D facade benchmark todate, comprising centimeter-accurate point clouds with hierarchical, har-monized, and architecturally grounded semantic labels totaling 2.7 bil-lion annotated points. Through a systematic evaluation of representativepoint-, graph- and transformer-based architectures, we show that cur-rent methods struggle to recognize fine-grained architectural elementsand degrade significantly across geographic regions, with the best mod-els achieving only up to 33 IoU on the fine-grained LoFG3 benchmark.By combining geometric precision with standardized semantics at un-precedented scale, UnderOneFacade establishes a rigorous benchmarkfor developing robust and transferable 3D segmentation models for andbeyond facade understanding. The dataset, evaluation scripts, and pre-trained models are available here: https://jiangyuanwangyi.github.io/UnderOneFacade_official/