Neural implicit functions have become popular for representing surfaces because they offer an adaptive resolution and support arbitrary topologies. While previous works rely on ground truth point clouds, they often ignore the effect of input quality and sampling methods on the reconstruction. In this paper, we introduce NeuroSURF, which generates significantly improved qualitative and quantitative reconstructions driven by a novel sampling and interpolation technique. We show that employing a sampling technique that considers the geometric characteristics of inputs can enhance the training process. To this end, we introduce a strategy that efficiently computes differentiable geometric features, namely, mean curvatures, to augment the sampling phase during the training period. Moreover, we augment the neural implicit surface representation with uncertainty, which offers insights into the occupancy and reliability of the output signed distance value, thereby expanding representation capabilities into open surfaces. Finally, we demonstrate that NeuroSURF leads to state-of-the-art reconstructions on both synthetic and real-world data.

翻译：神经隐式函数因能够提供自适应分辨率并支持任意拓扑结构而成为表面表示的主流方法。然而，以往研究依赖于真实点云，却常忽略输入质量与采样方法对重建效果的影响。本文提出的NeuroSURF方法通过新型采样与插值技术，实现了显著改善的定性与定量重建结果。我们论证了考虑输入几何特征的采样技术能够增强训练过程。为此，我们引入了一种策略，可在训练期间高效计算可微几何特征（即平均曲率）以增强采样阶段。此外，我们通过不确定性机制增强神经隐式表面表示，从而获得输出有符号距离值的占据状态与可靠性信息，进而将表示能力扩展至开放表面。最后，实验证明NeuroSURF在合成数据与真实数据上均达到了最先进的性能水平。