本期带来上周在arXiv公开的论文13篇，涉及3D重建、3D生成与编辑。一、3D重建1. [2024-arXiv] Single-View 3D Human Digitalization with Large Reconstruction Models作者机构：Zhenzhen Weng, et al. Stanford University, Adobe Research.论文地址：https://arxiv.org/pdf/2401.12175.pdf项目地址：https://zzweng.github.io/humanlrm贡献：We introduce Human-LRM, a specialized LRM for humans with improved surface fidelity. Being trained on an extensive dataset (more than 10K shapes) with both multiview RGB data and 3D scans, our model attains substantially enhanced generalizability and excels across a wider spectrum of scenarios and applications.To enhance the applicability of Human-LRM for in-thewild scenarios, we distills multi-view Human-LRM into single-view through a conditional diffusion model. The generative Human-LRM enables better handling of unseen body parts from both other views and occlusions.Through extensive experiments, we show that HumanLRM outper
………………………………
