Schelling points on 3D surface meshes

Abstract

This paper investigates "Schelling points" on 3D meshes, feature points selected by people in a pure coordination game due to their salience. To collect data for this investigation, we designed an online experiment that asked people to select points on 3D surfaces that they expect will be selected by other people. We then analyzed properties of the selected points, finding that: 1) Schelling point sets are usually highly symmetric, and 2) local curvature properties (e.g., Gauss curvature) are most helpful for identifying obvious Schelling points (tips of protrusions), but 3) global properties (e.g., segment centeredness, proximity to a symmetry axis, etc.) are required to explain more subtle features. Based on these observations, we use regression analysis to combine multiple properties into an analytical model that predicts where Schelling points are likely to be on new meshes. We find that this model benefits from a variety of surface properties, particularly when training data comes from examples in the same object class.

New Interest Points Detector for 3D Objects Recognition

“…There is no exact definition for interest points up to now and different applications may have different requirements for these points. For example, some applications are inclined to choose the interest points with high repeatability or prefer them with a higher discriminability and also others may require them with semantic information, such as Schelling point [11]. Thus, when designing the interest point detection algorithms, we usually need to define the proper interest point response function according to the practical situation.…”

A salience measure for 3D shape decomposition and sub-parts classification

“…Nouri et al compute local descriptors on each vertex and then a measure of saliency based on a weighted average at different scales [12]. These methods usually compare their results with outcome of user studies, that either captured eye fixations of users watching 3d objects [13], or matched manually designated interest points [14].…”

A Unified Deep Metric Representation for Mesh Saliency Detection and Non-Rigid Shape Matching

“…The first task we are interested in is saliency detection, which aims to compute a saliency map for an input mesh that signifies the perceptual or semantic importance of surface regions [1], [3]. Despite highlighting semantically important Fig.…”

“…regions, saliency maps are also found to be consistent on surfaces of the same object category [3]. However, the intracategory consistency of saliency has been ignored by previous saliency detection methods [1], [3]- [5], which limits their generalization abilities under complex intra-category shape deformations.…”

“…The other task we focus on is non-rigid shape matching, which finds semantically meaningful surface correspondences across meshes irrespective of the shape deformations among them [2]. As found in [3], human annotators tend to agree on a consistent set of semantically important regions on surfaces of the same object category, without communicating with each other during annotation. This shows that saliency is a strong deformation-invariant cue of shape matching within a category.…”