Generating 3D visual scenes is at the forefront of visual generative AI, but current 3D generation techniques struggle with generating scenes with multiple high-resolution objects. Here we introduce Lay-A-Scene, which solves the task of Open-set 3D Object Arrangement, effectively arranging unseen objects. Given a set of 3D objects, the task is to find a plausible arrangement of these objects in a scene. We address this task by leveraging pre-trained text-to-image models. We personalize the model and explain how to generate images of a scene that contains multiple predefined objects without neglecting any of them. Then, we describe how to infer the 3D poses and arrangement of objects from a 2D generated image by finding a consistent projection of objects onto the 2D scene. We evaluate the quality of Lay-A-Scene using 3D objects from Objaverse and human raters and find that it often generates coherent and feasible 3D object arrangements.
@article{arxiv.2406.00687,
title = {Lay-A-Scene: Personalized 3D Object Arrangement Using Text-to-Image Priors},
author = {Ohad Rahamim and Hilit Segev and Idan Achituve and Yuval Atzmon and Yoni Kasten and Gal Chechik},
journal= {arXiv preprint arXiv:2406.00687},
year = {2024}
}