Generating geometric 3d reconstructions from Neural Radiance Fields (NeRFs)
is of great interest. However, accurate and complete reconstructions based on
the density values are challenging. The network output dep
本文提出了一种使用预训练的 NeRF 模型对 3D 场景进行对齐的方法,该方法通过将传统的关键点检测和点集对齐的流程应用于 3D 密度场来实现。为了在 3D 中描述角点作为关键点,我们建议使用通用的预训练描述符生成神经网络。通过对比学习策略,可以方便地训练描述符网络。我们的方法作为全局方法可以有效地注册 NeRF 模型,从而使未来的大规模 NeRF 构建成为可能。
Neural Radiance Fields (NeRFs) are a new representation of 3D scenes for view synthesis and image-based rendering, widely used and extended by thousands of papers, with potential for future advancements in 3D representations.