Tencent Hunyuan's WorldRecon Reconstructs Scenes From Unposed Photos, Cuts ScanNet++ Chamfer Distance 43%
Summary
Zhejiang University and Tencent Hunyuan publish WorldRecon, a generative multi-view framework that rebuilds large-scale 3D scene meshes with PBR materials from unposed RGB images. It conditions block-by-block generation on local point clouds and inpaints seams, reducing Chamfer distance 43.1% and lifting F-score 12% over the strongest baseline on real ScanNet++ scenes.
Originally reported by arxiv.org
Read the original article →Original headline: Tencent Hunyuan's WorldRecon Reconstructs Scenes From Unposed Photos, Cuts ScanNet++ Chamfer Distance 43%