Lune

ICML2025

MIRROR: Make Your Object-Level Multi-View Generation More Consistent with Training-Free Rectification

Tianchi Xing, Bonan Li, Congying Han, Xinmin Qiu, Zicheng Zhang, Tiande Guo

出版方

摘要

More consistent Figure 1: Generated multi-view images by applying MIRROR on SyncDreamer (Liu et al., 2023b) and VideoMV (Zuo et al., 2024) for rectification. MIRROR effectively improves the consistency of generated results in both image-based and text-based multi-view generation models.