LDPoly
Source code for "LDPoly: Latent diffusion for polygonal road outline extraction in large-scale topographic mapping"
Description
This software contains the official implementation of LDPoly, a novel diffusion-model-based framework specifically designed for polygonal road outline extraction from high-resolution aerial imagery. It employs a dual-latent diffusion model that jointly diffuses and denoises road masks and keypoint heatmaps, which are subsequently vectorized into polygonal road outlines. The repository includes model implementation, training and inference configurations, polygonization procedures, evaluation scripts, and instructions for use. This archived version corresponds to the code associated with the ISPRS 2025 paper “LDPoly: Latent diffusion for polygonal road outline extraction in large-scale topographic mapping.”
- MIT
Reference papers
Mentions
- 1.Author(s): Xiaolong Zuo, Claudio Persello, Weiqin Jiao, Yu Wang, Zhenfeng ShaoPublished in ISPRS Journal of Photogrammetry and Remote Sensing by Elsevier BV in 2026, page: 203-22210.1016/j.isprsjprs.2026.05.049
- 2.Author(s): Boaz Mwubahimana, Dingruibo Miao, Yan Jianguo, Le Ma, Remy Dukundane, Xiao Huang, Swalpa Kumar Roy, Ruisheng WangPublished in IEEE Transactions on Geoscience and Remote Sensing by Institute of Electrical and Electronics Engineers (IEEE) in 2026, page: 5630917-563091710.1109/tgrs.2026.3707401
- 3.Author(s): Xiaoshan Yong, Jianyong Zhang, Yu Zhao, Qian Cui, Shijie Qiao, Yanjie Liu, Yugang Cao, Wu XiaoPublished in Remote Sensing by MDPI AG in 2026, page: 136710.3390/rs18091367
- 4.Author(s): Ligao Deng, Zhihao Xi, Jiansheng Chen, Yuhuan Ren, Jingbo Chen, Anzhi Yue, Yupeng Deng, Yu MengPublished in 202610.1016/j.isprsjprs.2026.10.017