PG-NeuS:对于多视图神经表面重建的强大和有效的点指导
IEEE transactions on visualization and computer graphics
|March 3, 2025
概括
通过模拟点云不确定性和使用神经投影进行精确指导,PG-NeuS增强了神经表面重建. 这种以点为导向的方法实现了显著的准确性和速度改进,同时保持了对杂数据的稳定性.
科学领域:
- 计算机视觉 计算机视觉
- 三维重建的3D重建
- 深度学习 (Deep Learning) 是一种深度学习.
背景情况:
- 多视图神经表面重建方法经常受到有限的准确性和高时间复杂性的困扰.
- 现有的方法难以充分利用先前信息,对数据扰动敏感,导致几何扭曲.
研究的目的:
- 提出一种新的点导向方法 (PG-NeuS),用于准确,高效和强大的神经表面重建.
- 解决当前重建技术中噪音,准确性和效率的挑战.
主要方法:
- 模拟点云的随机不确定性,以估计点可靠性和增强噪声强度.
- 引入神经投影模块来连接点和图像,提供精确点指导的几何约束.
- 设计一个Bias网络以补偿几何偏差,并通过使用高保真度点信息增强细节表示.
主要成果:
- 在DTU数据集上,PG-NeuS实现了11倍的速度提升和33.3%的精度提升.
- 该方法证明了高质量的表面重建,提高了效率,特别是在细粒度细节和光滑区域.
- 对噪音和稀疏的输入数据表现出强大的稳定性.
结论:
- 在神经表面重建方面,PG-NeuS提供了显著的进步,提供了最先进的性能.
- 拟议的方法有效地处理噪音数据,并提高重建效率和准确性.
- PG-NeuS是计算机视觉中高保真度3D表面生成的一个有前途的方法.
相关概念视频
02:06Newman Projections
Different notations are used to represent the three-dimensional structure of molecules on two-dimensional surfaces. One of the most commonly used representations is the dash-wedge formula. The dashed wedges, solid wedges, and the plane lines indicate the groups situated behind the plane, coming out of the plane, and in the plane, respectively.
The organic molecules rotate across the single bonds leading to numerous temporary three-dimensional structures of varying energy known as conformers.
The organic molecules rotate across the single bonds leading to numerous temporary three-dimensional structures of varying energy known as conformers.
01:30Parametric Surfaces
A parametric surface in three-dimensional space is defined through a vector-valued function\begin{equation*}\mathbf{r}(u, v) = x(u, v)\mathbf{i} + y(u, v)\mathbf{j} + z(u, v)\mathbf{k}\end{equation*}where u and v are parameters within a specified domain D in the uv-plane. The functions x(u, v), y(u, v), and z(u, v) define the coordinates of points on the surface. As u and v vary over D, the position vector r(u, v) traces a continuous surface in space. This parametric representation is essential...