文章预览
本期带来上周在arXiv公开的论文6篇,涉及3D重建与生成、3D编辑。 一、3D重建与生成 1.[2024-arXiv] Self-augmented Gaussian Splatting with Structure-aware Masks for Sparse-view 3D Reconstruction 作者机构:Lingbei Meng, et al. Peking University 论文地址: https://arxiv.org/pdf/2408.04831 贡献: We propose a novel coarse-to-fine paradigm for 3D Gaussian splatting model with both 3D geometry augmentation and perceptual view augmentation, aiming at enhancing consistent and detailed representation for sparse-view 3D reconstruction. To mitigate the issues that 3D Gaussians still suffer in regions with occlusion and unobserved regions, we have developed an effective structure-aware mask strategy. We apply point-based masks and patch-based masks on the coarse and fine Gaussians to further improve the model’s robustness against sparse inputs and noise, respectively. Experimental results show that the proposed method outperforms current state-of-the-art me
………………………………