In contrast, our formulation simultaneously infers a single-step noise as the continuous quantity and the final 2D coordinate as the discrete quantity, achieving superior generation capabilities (See Sect. However, we found that these pure discrete representations do not train well, probably because the diffusion process is continuous in nature. In this regard, several works use discrete state space or learn an embedding of discrete data in the DM formulation. What makes our task unique and challenging is the precise geometric incident relationships, such as parallelism, orthogonality, and corner-sharing among different components, which continuous coordinate regression would never achieve. Molecular Conformation Generation and 3D shape generation The key difference is that HouseDiffusion processes a vector geometry representation from start to finish, and hence, directly generating vector floorplan samples. Our paper also tackles a graph-constrained floorplan generation with a bubble diagram as the constraint. use graph-theoretic and linear optimization techniques to generate floorplans. Although their method generates vector floorplans directly, it is limited to rectangular shapes.Īlong with the adjacency graph as the input, Yin et al. Given a set of room types and their area sizes as the constraint, Luo and Huang proposed a vector generator and a raster discriminator to train a GAN model using differential rendering. proposed to iteratively generate connectivity graphs of rooms and a floorplan semantic segmentation mask. proposed Graph2Plan that retrieves a graph layout from a dataset and generates room bounding boxes as well as a floorplan in an ad-hoc way. used the embedded input boundary as an additional input feature to predict a floorplan. Given the boundary of a floorplan, Upadhyay et al. The authors further improved the quality of the generation by House-GAN++, which iteratively refines a layout. House-GAN generates segmentation masks of different rooms and combines them to a single floorplan. Proposed House-GAN as a graph constrained floorplan generative model via Generative Adversarial Network. The research area has further flourished with the emergence of deep learning. Generation of 3D buildings and floorplans has been an active area of research from a pre- deep learning era. One could use a discrete representation such as one hot encoding over possible coordinate values with classification, but this causes a label imbalance (i.e., most values are 0 in the encoding) and fails the network training. A wall might be shared with adjacent rooms further.ĭirect regression of 2D coordinates would never achieve these relationships. For example, a wall is usually axis-aligned, where the coordinate values of adjacent corners are exactly equal. On the other hand, direct generation of vector floorplans is not trivial either.ĭifferent from the generation of images or natural languages, structured geometry exhibit precise incident relationships among architectural components (e.g., doors and rooms). The raster analysis is good at local shape refinement but lacks in global reasoning, and requires non-trivial post-processing for vectorization The issue is the raster geometry analysis via convolutions, where a room is represented as a binary image. Website with supplementary video and document is hereĭespite recent progress, state-of-the-art floorplan generative models produce samples that are incompatible with the input constraint, lack in variations, or do not look like floorplans. Structures and controlling the exact number of corners per room. With significant margins, while being capable of generating non-Manhattan Makes significant improvements in all the metrics against the state-of-the-art We have evaluated our approach on RPLAN dataset. Generates vector-graphics floorplans via a discrete and continuous denoising Our diffusion model employs a Transformer architecture at the core, whichĬontrols the attention masks based on the input graph-constraint and directly We represent aįloorplan as 1D polygonal loops, each of which corresponds to a room or a door. Our task is graph-conditionedįloorplan generation, a common workflow in floorplan design. Parallelism, orthogonality, and corner-sharing. Precisely invert the continuous forward process and 2) the final 2D coordinateĪs the discrete quantity to establish geometric incident relationships such as Inference objectives: 1) a single-step noise as the continuous quantity to The paper presents a novel approach for vector-floorplan generation via aĭiffusion model, which denoises 2D coordinates of room/door corners with two
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |