SceneFunctioner: Tailoring Large Language Model for Function-Oriented Interactive Scene Synthesis
Keywords: Scene Synthesis, Multi-Function Design, Large Language Model, User Interaction
TL;DR: We introduce SceneFunctioner, a three-step interactive framework that tailors the LLM to assist users in designing multi-functional scenes.
Abstract: With the Large Language Model (LLM) skyrocketing in recent years, an increasing body of research has focused on leveraging these models for 3D scene synthesis. However, most existing works do not emphasize homeowner's functional preferences, often resulting in scenes that are logically arranged but fall short of serving practical functions. To address this gap, we introduce SceneFunctioner, an interactive scene synthesis framework that tailors the LLM to prioritize functional requirements. The framework is interactive, enabling users to select functions and room shapes. SceneFunctioner first distributes these selected functions into separate areas called zones and determines the furniture for each zone. It then organizes the furniture into groups before arranging them within their respective zones to complete the scene design. Quantitative analyses and user studies showcase our framework’s state-of-the-art performance in terms of both design quality and functional consistency with the user input.
Supplementary Material: zip
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: 908
Loading