SHREC'19 (SHREC'19 track Matching Humans with Different Connectivity)

形状匹配在几何处理和形状分析中起着重要作用。在过去的几十年中，许多研究致力于提高表面之间的匹配质量。这项巨大的努力是由多个应用程序推动的，例如对象检索、动画和信息传输，仅举几例。形状匹配通常分为两大类：刚性匹配和非刚性匹配。在这两种情况下，标准评估通常在共享相同连通性的形状上执行，换句话说，由相同网格表示的形状。这主要是由于为这些形状提供了“自然”的基本事实。事实上，在大多数情况下，一致的连通性直接导致顶点之间的地面实况对应。然而，这个标准实践显然不允许估计方法对不同连接的鲁棒性。通过这条赛道，我们提出了一个基准来评估当要匹配的形状具有不同的连通性时点对点匹配管道的性能（参见图 1）。我们考虑同时存在 1) 不同的网格划分，2) 3D 空间中的刚性变换，3) 非刚性变形，4) 不同的顶点密度，范围从 5K 到超过 50K，以及 5) 由网格粘合引起的拓扑变化接触领域。这些形状之间的对应关系是通过最近提出的配准管道 FARM [1] 获得的。该方法为来自不同数据集的大量人体网格提供了 SMPL 模型 [2] 的高质量配准，我们从中获得了所有已注册网格和 SMPL 本身的明确对应关系。

Shape matching plays a critical role in geometric processing and shape analysis. Over the past few decades, extensive research has been dedicated to improving the quality of surface matching. This substantial research effort is driven by a wide range of applications, such as object retrieval, animation, and information transfer, to name just a few. Shape matching is generally categorized into two broad classes: rigid matching and non-rigid matching. In both cases, standard evaluations are typically conducted on shapes that share identical connectivity, in other words, shapes represented by the same mesh. This is primarily because such shapes provide a "natural" ground truth. In fact, in most cases, consistent connectivity directly yields ground-truth correspondences between vertices. However, this standard practice clearly fails to enable estimation methods to be evaluated for their robustness against varying connectivity. Through this track, we propose a benchmark to evaluate the performance of point-to-point matching pipelines when the shapes to be matched have different connectivity (see Figure 1). We consider scenarios with the following coexisting factors: 1) varying mesh discretizations, 2) rigid transformations in 3D space, 3) non-rigid deformations, 4) different vertex densities ranging from 5K to over 50K, and 5) topological changes in contact regions induced by mesh gluing. The correspondences between these shapes are obtained via the recently proposed registration pipeline FARM [1]. This method achieves high-quality registration of the SMPL model [2] for a large number of human meshes from different datasets, based on which we derive explicit correspondences between all registered meshes and the SMPL model itself.

搜集汇总

背景与挑战

背景概述

SHREC'19是一个专注于评估不同连通性人体形状匹配性能的数据集，包含多种复杂变形因素和拓扑变化，适用于3D形状匹配和分析研究。数据集通过FARM配准管道提供高质量的人体网格对应关系，支持从5K到超过50K顶点密度的网格。