Answers are grounded in this paper and cite their sources.
Textured 3D building models are increasingly used as visual and spatial assets in urban digital twins, yet their fragmented texture-atlas layouts make localized and structure-preserving disaster appearance adaptation difficult. This study proposes a heatmap-guided facade fire appearance stylization framework that modifies building textures without changing the original mesh geometry, texture coordinates, or material configuration. The framework first rectifies fragmented atlas regions into facade-aligned images while preserving triangle-wise correspondences with the source atlas. Wall--window masks and edge-based structural guidance are then combined with a normalized fire heatmap that controls the spatial extent of the fire appearance. The modified facade images are subsequently reconstructed into the original texture atlas through piecewise-affine mapping. Experiments were conducted on 1,435 upright facade images extracted from 428 real building models. As the normalized appearance scale increased from 0.25 to 1.00, Target-LPIPS increased from 0.1708 to 0.2925. Full-dataset ablation showed that heatmap guidance substantially reduced unintended changes in non-wall regions and improved structural preservation. Compared with four representative diffusion-based stylization methods, the proposed method achieved the lowest Protected-LPIPS of 0.0816 and the highest Edge-SSIM of 0.6999. These results demonstrate controllable fire-appearance modification while preserving non-target facade content and architectural structure. Within this scope, the framework is intended as an AI-driven geovisualization approach for visual scenario generation that complements, rather than replaces, physics-based fire simulation.
No takes yet. Share an insight, caveat, or question.
Wi et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: