Tokenization Allows Multimodal Large Language Models to Understand, Generate and Edit Architectural Floor Plans

研究简介

建筑平面图设计需要对几何、语义和空间层次进行联合推理，这对当前的AI系统来说仍然是一个重大挑战。HouseMind是一个多模态大语言模型，在一个框架中统一了建筑平面图的理解、生成和编辑。它通过VQ-VAE引入离散房间实例token，将布局几何与符号推理连接起来，实现可控且可解释的操作。实验表明，该模型在保持高效和可本地部署的同时，展现了优异的几何有效性和可控性。

引用格式

QIN S Z, WEBER R E, LU X Z. Tokenization Allows Multimodal Large Language Models to Understand, Generate and Edit Architectural Floor Plans[C/OL]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2026: 10430–10440. https://arxiv.org/abs/2603.11640.

BibTeX

@inproceedings{20260316HouseMind,
  title = {Tokenization Allows Multimodal Large Language Models to Understand, Generate and Edit Architectural Floor Plans},
  author = {Qin, Sizhong and Weber, Ramon Elias and Lu, Xinzheng},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2026},
  pages = {10430--10440}
}