CVPR2025
MOVIS: Enhancing Multi-Object Novel View Synthesis for Indoor Scenes
Ruijie Lu, Yixin Chen, Junfeng Ni, Baoxiong Jia, Yu Liu, Diwen Wan, Gang Zeng, Siyuan Huang
Abstract
Left 180° Down 5° Left 180° Down 5° Left 180° Down 5° Right 145° Up 10° Right 80° Left 50° Right 100° Up 5° Right 100° Up 5° Right 100° Up 5°O bjaverse SUNRGB-D 3D-FRONT NVS Cross View Matching Ours Zero-1-to-3 Ground Truth Figure 1 . Novel view synthesis and cross-view image matching. The first row shows that MOVIS generalizes to different datasets on novel view synthesis (NVS). We also show visualizations of cross-view consistency compared with Zero-1-to-3 [8] and ground truth by applying image-matching. MOVIS can match a significantly greater number of points, closely aligned with the ground truth.