CF-GISS: Collision-Free Generative 3D Indoor Scene Synthesis with Controllable Floor Plans and Optimized Layouts

27 Sept 2024 (modified: 16 Feb 2025)Submitted to ICLR 2025EveryoneRevisionsBibTeXCC BY 4.0
Keywords: Indoor Scene Synthesis, 3D scene generation, Procedural generation, Generative models
TL;DR: A novel framework for generative 3D indoor scene synthesis that ensures collision-free layouts while conditioning on controllable floor plans, along with a novel indoor dataset
Abstract:

We introduce CF-GISS, a novel framework for generative 3D indoor scene synthesis that ensures collision-free scene layouts by incorporating an image-based intermediate layout representation. In contrast to existing methods that directly construct the scene graph or object list, our approach facilitates substantially more effective prevention of collision artifacts as out-of-distribution (OOD) scenarios during generation. Furthermore, CF-GISS conditions layout generation on floor plans controllable via images or textual descriptions, enabling the production of coherent, house-wide layouts that are robust to variations in geometric and semantic structures. Our framework demonstrates state-of-the-art performance on the 3D-FRONT dataset, delivering high-quality, collision-free scene synthesis while offering flexibility in accommodating a range of floor plan structures. Additionally, we propose a novel dataset with significantly expanded coverage of household items and room configurations, as well as improved data quality.

Primary Area: applications to computer vision, audio, language, and other modalities
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.
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: 9376
Loading

OpenReview is a long-term project to advance science through improved peer review with legal nonprofit status. We gratefully acknowledge the support of the OpenReview Sponsors. © 2025 OpenReview