TLC-Plan: A Two-Level Codebook Based Network for End-to-End Vector Floorplan Generation

2026-02-06

AI Summary

Existing automatic floorplan generation methods typically operate in raster space and rely on post-hoc vectorization, leading to structural inconsistencies and hindering end-to-end training. This work proposes the first end-to-end vector-based floorplan generation model, which employs a two-level VQ-VAE to construct a hierarchical codebook that separately captures global layout and local geometry. A unified CodeTree representation encodes both room bounding boxes and polygon-level semantics. Leveraging an autoregressive Transformer, the model directly generates diverse, topologically valid designs from boundary inputs without requiring explicit topological or dimensional priors, aligning closely with human design workflows. It achieves state-of-the-art performance on both the RPLAN (FID=1.84, MSE=2.06) and LIFULL datasets, significantly improving generation quality and practical utility.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Automated floorplan generation aims to improve design quality, architectural efficiency, and sustainability by jointly modeling global spatial organization and precise geometric detail. However, existing approaches operate in raster space and rely on post hoc vectorization, which introduces structural inconsistencies and hinders end-to-end learning. Motivated by compositional spatial reasoning, we propose TLC-Plan, a hierarchical generative model that directly synthesizes vector floorplans from input boundaries, aligning with human architectural workflows based on modular and reusable patterns. TLC-Plan employs a two-level VQ-VAE to encode global layouts as semantically labeled room bounding boxes and to refine local geometries using polygon-level codes. This hierarchy is unified in a CodeTree representation, while an autoregressive transformer samples codes conditioned on the boundary to generate diverse and topologically valid designs, without requiring explicit room topology or dimensional priors. Extensive experiments show state-of-the-art performance on RPLAN dataset (FID = 1.84, MSE = 2.06) and leading results on LIFULL dataset. The proposed framework advances constraint-aware and scalable vector floorplan generation for real-world architectural applications. Source code and trained models are released at https://github.com/rosolose/TLC-PLAN.

Problem

• floorplan generation
• vector representation
• end-to-end learning
• structural consistency
• architectural design

Innovation

• vector floorplan generation
• two-level codebook
• VQ-VAE
• CodeTree
• autoregressive transformer