Graph machine learning using 3D topological models

Classification of urban and architectural works using machine learning techniques have typically focused on 2D pixel-based image recognition. In this paper we present a novel proof-of-concept workflow that enables a machine learning computer system to learn to classify 3D conceptual models based on topological graphs rather than 2D images. The system leverages two main technologies. The first is a custom designed software library that enhances the representation of 3D models through non-manifold topology and embedded semantic information. The second is an end-to-end deep graph convolutional network (DGCNN) that accepts graphs of an arbitrary structure without the need to first convert them into vectors. The experimental workflow consists of two stages. In the first stage, a generative parametric system was designed to create a large synthetic dataset of an urban block with several typological categories. The geometric models were then automatically labelled and converted into semantically rich topological dual graphs. In the second stage, the dual graphs were imported into the DGCNN for graph classification. Experiments demonstrate that the proposed workflow achieves accuracy results that are highly competitive with DGCNN’s classification of benchmark graph datasets.

[1]  R. Aish,et al.  Topologic - A toolkit for spatial and topological modelling , 2018, Proceedings of the 36th International Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe) [Volume 2].

[2]  Max Welling,et al.  Semi-Supervised Classification with Graph Convolutional Networks , 2016, ICLR.

[3]  Tf Thomas Krijnen,et al.  Assessing implicit knowledge in BIM models with machine learning , 2015 .

[4]  Jeffrey C Grossman,et al.  Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties. , 2017, Physical review letters.

[5]  Frank Brown,et al.  A Classification of Built Forms , 2000 .

[6]  Lutz Plümer,et al.  CityGML – Interoperable semantic 3D city models , 2012 .

[7]  Christian Derix,et al.  Digital intuition – Autonomous classifiers for spatial analysis and empirical design , 2014 .

[8]  Kathryn Fraughnaugh,et al.  Introduction to graph theory , 1973, Mathematical Gazette.

[9]  Yann LeCun,et al.  Spectral Networks and Deep Locally Connected Networks on Graphs , 2014 .

[10]  Zhanxing Zhu,et al.  Spatio-temporal Graph Convolutional Neural Network: A Deep Learning Framework for Traffic Forecasting , 2017, IJCAI.

[11]  Hamidreza Kasaei,et al.  OrthographicNet: A Deep Learning Approach for 3D Object Recognition in Open-Ended Domains , 2019, ArXiv.

[12]  B. Hillier,et al.  The Social Logic of Space , 1984 .

[13]  Jorge Gil,et al.  On the discovery of urban typologies: data mining the many dimensions of urban form , 2011, Urban Morphology.

[14]  Luc De Raedt,et al.  Graph Invariant Kernels , 2015, IJCAI.

[15]  Yixin Chen,et al.  An End-to-End Deep Learning Architecture for Graph Classification , 2018, AAAI.

[16]  Didier Stricker,et al.  Trained 3D Models for CNN based Object Recognition , 2017, VISIGRAPP.

[17]  S. V. N. Vishwanathan,et al.  Graph kernels , 2007 .

[18]  Nils M. Kriege,et al.  Subgraph Matching Kernels for Attributed Graphs , 2012, ICML.

[19]  Wassim Jabi,et al.  Topologic: Tools to explore architectural topology , 2018, AAG.

[20]  Qiang Yang,et al.  Bike flow prediction with multi-graph convolutional networks , 2018, SIGSPATIAL/GIS.

[21]  Cyrus Shahabi,et al.  Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting , 2017, ICLR.

[22]  Jakob Beetz,et al.  A scalable network of concept libraries using distributed graph databases , 2014 .