Continuous generalization is a key technology for achieving continuous-scale representation of map features, with continuous building simplification being an important component. Existing methods often suffer from issues such as misalignment of feature points, disruption of local similarity, and a lack of hierarchical simplification. To address these challenges, this paper proposes a continuous building simplification model based on a conditional diffusion model. The model adopts an architecture comprising a PSRT (Pre-Trained Shape Feature Representations from Transformers) encoder and a BART (Bidirectional and Auto-Regressive Transformers) decoder, trained with sequences of building contour coordinate points as the target output and corresponding sequences of building control points as conditional input. After training, continuous building simplification is achieved by combining latent space interpolation with initial noise interpolation. Experimental results demonstrate that the proposed method can generate reasonable and coherent building simplification results at intermediate scales.
Chen et al. (Sun,) studied this question.