SPE-MVS：融合单目深度先验的空间位置编码增强多视图立体
SPE-MVS: Spatial Position Encoding Enhanced Multi-View Stereo with Monocular Depth Priors
把单目深度先验编码成统一度量空间下的三维空间位置编码，并与图像特征一起送入代价体构建，为弱纹理、反光等光度一致性失效区域提供几何先验；做 MVS 或深度估计时可复用这一『空间位置相似度』替代纯光度匹配的思路。
原文依据 The SPE represents the 3D positional information of pixels in each image within a unified metric space, constructed using monocular depth priors.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论基于学习的多视图立体（MVS）方法已成为该领域的主流，其依赖通过多视图特征相似度计算来构建代价体。然而，现有方法严重依赖跨视图的光度一致性，在困难区域表现不佳。为克服这一局限，我们提出 SPE-MVS，一种由空间位置编码（SPE）增强的新型 MVS 框架。SPE 在统一度量空间内表示每幅图像中像素的三维位置信息，并利用单目深度先验构建。我们将 SPE 与图像数据一同作为输入，并引入光度-空间混合特征提取器以及 SPE 增强的代价体构建模块。这些组件融入了基于空间位置的相似度计算，显著提升了在困难区域的鲁棒性。此外，我们提出单目深度引导增强（MDGE）模块，利用单目深度先验增强深度概率图，从而进一步提升深度估计性能。大量实验表明，我们的方法显著提升了困难区域的重建质量，并在多个基准上取得最优性能。代码将发布于 https://github.com/bdwsq1996/SPE-MVS。
英文原文摘要 1467 字符
Learning-based Multi-View Stereo (MVS) methods have become the mainstream in the field, relying on the construction of cost Learning-based Multi-View Stereo (MVS) methods have become the mainstream in the field, relying on the construction of cost volumes through multi-view feature similarity computation. However, existing methods depend heavily on photometric consistency across views, leading to poor performance in challenging regions. To overcome this limitation, we propose SPE-MVS, a novel MVS framework enhanced with Spatial Position Encoding (SPE). The SPE represents the 3D positional information of pixels in each image within a unified metric space, constructed using monocular depth priors. We integrate the SPE alongside image data as input and introduce a Photometric-Spatial Hybrid Feature Extractor, along with an SPE-enhanced cost volume construction module. These components incorporate spatial position-based similarity computation, substantially improving robustness in challenging areas. Furthermore, we propose a Monocular Depth-guided Enhancement (MDGE) module that enhances depth probability map using monocular depth priors, thereby further boosting the depth estimation performance. Extensive experiments demonstrate that our method significantly improves reconstruction quality in difficult regions and achieves state-of-the-art (SOTA) performance on multiple benchmarks. The code will be released at https://github.com/bdwsq1996/SPE-MVS.