Computer-Aided Layout Generation for Building Design: A Review

TL;DR

This survey addresses the problem of computer-aided building layout generation by organizing methods into three core areas: residential floorplan generation, scene layout synthesis, and broader building-site layouts. It contrasts traditional architecture-driven optimization with modern data-driven approaches, emphasizing representations (raster vs vector) and user-input conditioning (boundary constraints and bubble diagrams). The paper catalogs leading benchmark datasets (e.g., RPLAN, LIFULL, Structured3D, Zillow, CubiCasa5k, 3D-FRONT, 3DSSG, SG-FRONT) and evaluation metrics (Realism, FID/KID, diversity, compatibility, graph-constraint metrics, boundary IoU), and summarizes current state-of-the-art models, including diffusion-based and graph/transformer architectures. It identifies open problems such as limited input modalities, scalability to multi-room or urban-scale layouts, end-to-end diversity, and generalization across datasets, and suggests future directions toward multi-modal foundations and integration with large language models to enhance practical CAD workflows.

Abstract

Generating realistic building layouts for automatic building design has been studied in both the computer vision and architecture domains. Traditional approaches from the architecture domain, which are based on optimization techniques or heuristic design guidelines, can synthesize desirable layouts, but usually require post-processing and involve human interaction in the design pipeline, making them costly and timeconsuming. The advent of deep generative models has significantly improved the fidelity and diversity of the generated architecture layouts, reducing the workload by designers and making the process much more efficient. In this paper, we conduct a comprehensive review of three major research topics of architecture layout design and generation: floorplan layout generation, scene layout synthesis, and generation of some other formats of building layouts. For each topic, we present an overview of the leading paradigms, categorized either by research domains (architecture or machine learning) or by user input conditions or constraints. We then introduce the commonly-adopted benchmark datasets that are used to verify the effectiveness of the methods, as well as the corresponding evaluation metrics. Finally, we identify the well-solved problems and limitations of existing approaches, then propose new perspectives as promising directions for future research in this important research area. A project associated with this survey to maintain the resources is available at awesome-building-layout-generation.

Figures (14)

• Figure 1: The diverse building layout types used in mainstream computer-aided architectural design methods. The floorplan and scene furniture layouts are the two most common types of layouts. Moreover, there exist other types including roofs, building volumes, community or urban-level layout, etc.
• Figure 2: The typical workflow of different representative deep generative models, including GANs, VAES, Autoregressive Models and Diffusion Models (DMs).
• Figure 3: A generic learning-based floorplan generation pipeline with user input. We show two major user input conditions, residential boundary and bubble diagram--commonly used in current studies. Methods using rasterized representation aim to generate a set of room masks then perform post-processing and integrate the outputs into a vectorized floorplan. For vectorized methods, the outcomes can be directly integrated into a floorplan. The icons of the boundary input and the final generation are referred from the diagrams used in RPLAN.
• Figure 4: HouseGAN is a representative floorplan generation approach using the rasterized representation. It first parses the input bubble diagram, then generates the room masks separately with a generator and discriminator architecture. The separately generated rooms are subsequently integrated together and post-processed to finalize the floorplan design.
• Figure 5: A tree-structured diagram to illustrate the categorization of different approaches and representative methods.