CVPR2023

LEGO-Net: Learning Regular Rearrangements of Objects in Rooms

Qiuhong Anna Wei, Sijie Ding, Jeong Joon Park, Rahul Sajnani, Adrien Poulenard, Srinath Sridhar, Leonidas J. Guibas

摘要

Figure 1. LEGO-Net LEarns to reGularly rearrange Objects in a messy indoor scene via an iterative denoising process. Different from scene synthesis or methods that require goal state specification, our method learns clean re-arrangements directly from data, retains the flavor of the original scene, and minimizes object travel distance.