The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLIII-B2-2021
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLIII-B2-2021, 337–344, 2021
https://doi.org/10.5194/isprs-archives-XLIII-B2-2021-337-2021
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLIII-B2-2021, 337–344, 2021
https://doi.org/10.5194/isprs-archives-XLIII-B2-2021-337-2021

  28 Jun 2021

28 Jun 2021

SYNTHETIC DATA GENERATION PIPELINE FOR GEOMETRIC DEEP LEARNING IN ARCHITECTURE

S. Fedorova1, A. Tono2, M. S. Nigam3, J. Zhang4, A. Ahmadnia1, C. Bolognesi1, and D. Michels4 S. Fedorova et al.
  • 1Politecnico di Milano, Milan, Italy
  • 2CDI, Computational Design Institute, San Francisco, USA
  • 3IIIT Hyderabad, India
  • 4KAUST, Saudi Arabia

Keywords: Synthetic 3D Dataset, Procedural Generation, Dataset Generation Pipeline, Architecture, Geometric Deep Learning, 3D Reconstruction

Abstract. With the growing interest in deep learning algorithms and computational design in the architectural field, the need for large, accessible and diverse architectural datasets increases. Due to the complexity of such 3D datasets, the most widespread techniques of 3D scanning and manual building modeling are very time-consuming, which does not allow to have a sufficiently large open-source dataset. We decided to tackle this problem by constructing a field-specific synthetic data generation pipeline that generates an arbitrary amount of 3D data along with the associated 2D and 3D annotations. The variety of annotations, the flexibility to customize the generated building and dataset parameters make this framework suitable for multiple deep learning tasks, including geometric deep learning that requires direct 3D supervision. Creating our building data generation pipeline we leveraged the experts’ architectural knowledge in order to construct a framework that would be modular, extendable and would provide a sufficient amount of class-balanced data samples. Moreover, we purposefully involve the researcher in the dataset customization allowing the introduction of additional building components, material textures, building classes, number and type of annotations as well as the number of views per 3D model sample. In this way, the framework would satisfy different research requirements and would be adaptable to a large variety of tasks. All code and data is made publicly available: https://cdinstitute.github.io/Building-Dataset-Generator/.