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

Abstract

This paper proposes a novel graph-constrained generative adversarial network, whose generator and discriminator are built upon relational architecture. The main idea is to encode the constraint into the graph structure of its relational networks. We have demonstrated the proposed architecture for a new house layout generation problem, whose task is to take an architectural constraint as a graph (i.e., the number and types of rooms with their spatial adjacency) and produce a set of axis-aligned bounding boxes of rooms. We measure the quality of generated house layouts with the three metrics: the realism, the diversity, and the compatibility with the input graph constraint. Our qualitative and quantitative evaluations over 117,000 real floorplan images demonstrate that the proposed approach outperforms existing methods and baselines. We will publicly share all our code and data.

Room Classification on Floor Plan Graphs using Graph Neural Networks

“…We experiment with House-GAN dataset [18] that consists of 143,184 floor plans extracted in vector format from real floor plan images in Lifull Home's dataset [19]. There are 8 categories of rooms 1 .…”

In-game Residential Home Planning via Visual Context-aware Global Relation Learning

“…For example, (Wu et al 2019) proposes a two-stage method to iteratively locate rooms and walls given an input boundary while (Hu et al 2020) introduces an interactive solution in which users can specify some constraints during planning. In (Nauata et al 2020), they propose a convolutional message passing network named House-GAN that takes as input a bubble diagram and outputs the house layout with axisaligned bounding boxes. Unlike these tasks of indoor layout and floor plan design, our work focuses on outdoor home planning, and specifically, we aim to suggest locations for the new buildings.…”

“…Finally, our system outputs a location prediction indicates the "suitableness" for the next building unit. placement of a new component in an indoor scene automatically (Wang et al 2018Nauata et al 2020;Wu et al 2019). They use deep learning techniques e.g., the FiLM net (Perez et al 2018), to predict a component location as an attribute of the new node.…”

End-to-end Generative Floor-plan and Layout with Attributes and Relation Graph

“…This representation r i encodes the number of walls, doors, windows, and the distance among them in an adjacency matrix. This is similar to the representation used in HouseGAN [12]. The discriminator part d 1 determines whether the generated box of the floor-plan is real as illustrated in Figure 5.…”

“…In the conditional graphical generative module [12], the generation part g 2 takes a random vector z i and the representation r i of the ground truth floor-plan of the room as inputs. This representation r i encodes the number of walls, doors, windows, and the distance among them in an adjacency matrix.…”