SYNBUILD-3D: A multi-modal synthetic dataset of over 100,000 semantically enriched 3D building wireframes with AI-generated floor plans
Keywords: dataset, wireframe, 3D building, generative modeling, geometric deep learning
TL;DR: A multi-modal synthetic dataset of over 100,000 semantically enriched 3D building wireframes with AI-generated floor plans
Abstract: Modeling precise geometric and semantic relationships in 3D remains one of the greatest challenges in generative machine learning today, partly because of a lack of large 3D datasets in the public domain. Drawing upon the successful adoption of synthetic datasets in the computer vision community, we propose to address this challenge in the context of 3D buildings with SYNBUILD-3D, a large, multi-modal, and domain-specific dataset of more than 100,000 3D building wireframes along with their corresponding floor plan images. Unlike existing 3D building datasets, SYNBUILD-3D has been designed with and validated by building modeling and simulation experts, providing rich geometric and semantic information. As a result, SYNBUILD-3D is, to the best of our knowledge, the first 3D building dataset that provides interior and exterior building geometries, including the position and size of doors and windows derived from the floor plans. By releasing SYNBUILD-3D, we aim to offer the geometric deep learning community a high-quality dataset for conditional and unconditional 3D building generation tasks. In contrast to existing datasets that typically focus on modeling either the interior or exterior of 3D objects, SYNBUILD-3D can facilitate the development of generative algorithms that account for both perspectives while incorporating geometric and semantic constraints. The dataset and its associated codebase are available at GITHUB LINK.
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 13575
Loading