Exploring Architectural Design 3D Reconstruction Approaches through Deep Learning Methods: A Comprehensive Survey

Abstract

In recent years, substantial advancements in computer vision have been propelled by the advent of deep learning techniques. The transformative impact of deep learning is evident across various computer vision applications, spanning object detection, image classification, and semantic segmentation. Within the dynamic realm of architectural parametric design, a pivotal force shaping contemporary and future architectural pursuits, designers frequently encounter scenarios necessitating the reshaping, augmentation, or modification of building components and facades. This demand arises from diverse factors, including the integration of human-interactive facades, the implementation of sustainable facade systems, and more. In such instances, the foundational structure must possess malleability, enabling openness to edits. However, the grand scale and intricate details inherent in architectural designs pose a challenge, as a densely detailed mesh comprising millions of elements becomes increasingly unwieldy and challenging to modify for subsequent stages. This challenge underscores the critical need to generate objects capable of seamless evolution in tandem with evolving design requirements. This paper, will focus on categorizing 3D deep learning approaches into Learning-based 3D reconstruction in architecture and delving into their methods. Keywords: Algorithm Development, Computational Design, Learning-based Generative Design, Deep Learning, 3D Reconstruction

Topics & Concepts