Abstract:We present a novel graph Transformer generative adversarial network (GTGAN) to learn effective graph node relations in an end-to-end fashion for challenging graph-constrained architectural layout generation tasks. The proposed graph-Transformer-based generator includes a novel graph Transformer encoder that combines graph convolutions and self-attentions in a Transformer to model both local and global interactions across connected and non-connected graph nodes. Specifically, the proposed connected node attention (CNA) and non-connected node attention (NNA) aim to capture the global relations across connected nodes and non-connected nodes in the input graph, respectively. The proposed graph modeling block (GMB) aims to exploit local vertex interactions based on a house layout topology. Moreover, we propose a new node classification-based discriminator to preserve the high-level semantic and discriminative node features for different house components. To maintain the relative spatial relationships between ground truth and predicted graphs, we also propose a novel graph-based cycle-consistency loss. Finally, we propose a novel self-guided pre-training method for graph representation learning. This approach involves simultaneous masking of nodes and edges at an elevated mask ratio (i.e., 40%) and their subsequent reconstruction using an asymmetric graph-centric autoencoder architecture. This method markedly improves the model's learning proficiency and expediency. Experiments on three challenging graph-constrained architectural layout generation tasks (i.e., house layout generation, house roof generation, and building layout generation) with three public datasets demonstrate the effectiveness of the proposed method in terms of objective quantitative scores and subjective visual realism. New state-of-the-art results are established by large margins on these three tasks.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to generate realistic building layouts (such as floor plans, roofs and building layouts) that conform to the given graph - structure constraints. Specifically, the author aims to effectively capture the local and global relationships between graph nodes by combining Graph Convolutional Networks (GCNs) and Transformer models, so as to achieve high - quality building layout generation. ### Main Problem Description 1. **Limitations of Existing Methods**: - Existing methods based on Convolutional Neural Networks (CNNs) perform poorly in dealing with long - distance dependencies because they lack an understanding of long - range dependencies in the input graph. - Traditional GAN (Generative Adversarial Network) methods can generate realistic images, but they have limited effectiveness in tasks with graph - structure constraints. 2. **Objectives**: - Propose a new Graph Transformer GAN (GTGAN) to solve the building layout generation task under graph - structure constraints. - The generated layouts should not only look realistic but also maintain the consistency of high - level semantic features such as room type, size and position. ### Solutions To solve the above problems, the author proposes the following innovations: 1. **Graph Transformer Generator**: - Combine the advantages of Graph Convolutional Networks and Transformer to model local and global interactions simultaneously. - Introduce Connected Node Attention (CNA) and Non - Connected Node Attention (NNA) to capture the global relationships between connected and non - connected nodes in the graph respectively. - Use the Graph Modeling Block (GMB) to utilize local vertex interactions in the house layout topology. 2. **Node - Classification - Based Discriminator**: - The discriminator not only distinguishes between real and fake layouts but also classifies the generated rooms to preserve the high - order semantics and distinguishing features of different room types. 3. **Graph - Based Cycle - Consistency Loss**: - Propose a new Graph - Based Cycle - Consistency Loss to maintain the relative spatial relationships between the ground truth and the predicted graph. 4. **Self - Supervised Pre - training Method**: - Propose a new Graph Masked Modeling method. By randomly masking nodes and edges in the graph and using an asymmetric graph - centered auto - encoder architecture for reconstruction, the learning efficiency and effectiveness of the model are significantly improved. ### Experimental Verification Experiments on three challenging graph - constrained building layout generation tasks (house layout generation, house roof generation and building layout generation) prove the effectiveness and superiority of the proposed method. The experimental results show that GTGAN outperforms existing methods in both objective quantitative scores and subjective visual realism. ### Summary This paper solves the limitations of existing methods in generating building layouts under graph - structure constraints by introducing the Graph Transformer GAN framework, and achieves higher - quality and more realistic building layout generation.

Graph Transformer GANs with Graph Masked Modeling for Architectural Layout Generation

Graph Transformer GANs with Graph Masked Modeling for Architectural Layout Generation

Graph Transformer GANs for Graph-Constrained House Generation

GTA: Graph Transformer Adapter

House-GAN++: Generative Adversarial Layout Refinement Network towards Intelligent Computational Agent for Professional Architects

CRTGAN: Controllable Road Network Graphs Generation Via Transformer Based GAN

House-GAN: Relational Generative Adversarial Networks for Graph-constrained House Layout Generation

Roof-GAN: Learning to Generate Roof Geometry and Relations for Residential Houses

Adaptive Multi-Neighborhood Attention based Transformer for Graph Representation Learning

GraphGAN: Graph Representation Learning with Generative Adversarial Nets

Self-supervised Graph Representations with Generative Adversarial Learning

Graph Transformers for Large Graphs

Graph Transformer Networks Based Text Representation.

End‐to‐end generation of structural topology for complex architectural layouts with graph neural networks

Automated Layout Generation from Sites to Flats Using GAN and Transfer Learning

Graph Propagation Transformer for Graph Representation Learning

Generative Layout Modeling using Constraint Graphs

Exploring Task Unification in Graph Representation Learning via Generative Approach

PatchGT: Transformer over Non-trainable Clusters for Learning Graph Representations

EL-GAN: Edge-Enhanced Generative Adversarial Network for Layout-to-Image Generation

GTC: GNN-Transformer co-contrastive learning for self-supervised heterogeneous graph representation