MajutsuDataset
收藏资源简介:
MajutsuDataset是由中山大学构建的高质量多模态城市生成数据集,集成了二维语义布局与建筑高度图、多样化三维建筑资产以及专业级PBR材质与天空盒。该数据集包含数千个文本对齐的布局/高度图对,涵盖风格多样的几何约束三维资产,并配备精心策划的物理渲染材质库。数据集通过结构化流程构建,整合了语义分割、几何生成与材质合成技术,主要应用于三维城市生成领域,旨在解决语言驱动场景生成中结构一致性、风格多样性与材质真实性的核心挑战。
MajutsuDataset is a high-quality multimodal urban generation dataset constructed by Sun Yat-sen University. It integrates 2D semantic layouts and building height maps, diverse 3D architectural assets, as well as professional-grade PBR materials and skyboxes. This dataset contains thousands of text-aligned layout/height map pairs, covers geometrically constrained 3D assets with diverse styles, and is equipped with a carefully curated physically based rendering material library. Constructed through a structured workflow, the dataset integrates semantic segmentation, geometry generation, and material synthesis technologies. It is mainly applied in the field of 3D urban generation, aiming to address core challenges in language-driven scene generation: structural consistency, style diversity, and material authenticity.
MajutsuCity 数据集概述
数据集名称
MajutsuCity
核心描述
语言驱动的美学自适应城市生成,具有可控的3D资源和布局
数据集状态
- 待发布 MajutsuDataset
- 待发布 MajutsuCity
- 待发布布局生成模型
- 待发布材质纹理微调模型
- 待发布框架代码
- 待发布 MajutsuAgent 代码
相关资源
- 论文地址:https://arxiv.org/abs/2511.20415
- GitHub仓库:https://github.com/LongHZ140516/MajutsuCity
- 项目页面:https://longhz140516.github.io/MajutsuCity/
引用信息
bib @misc{huang2025majutsucitylanguagedrivenaestheticadaptivecity, title={MajutsuCity: Language-driven Aesthetic-adaptive City Generation with Controllable 3D Assets and Layouts}, author={Zilong Huang and Jun He and Xiaobin Huang and Ziyi Xiong and Yang Luo and Junyan Ye and Weijia Li and Yiping Chen and Ting Han}, year={2025}, eprint={2511.20415}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2511.20415}, }

- 1MajutsuCity: Language-driven Aesthetic-adaptive City Generation with Controllable 3D Assets and Layouts中山大学 · 2025年



