An evaluation of canonical forms for non-rigid 3D shape retrieval

Abstract

A B S T R A C TCanonical forms attempt to factor out a non-rigid shape's pose, giving a pose-neutral shape. This opens up the possibility of using methods originally designed for rigid shape retrieval for the task of non-rigid shape retrieval. We extend our recent benchmark for testing canonical form algorithms. Our new benchmark is used to evaluate a greater number of state-of-the-art canonical forms, on five recent non-rigid retrieval datasets, within two different retrieval frameworks. A total of fifteen different canonical form methods are compared. We find that the difference in retrieval accuracy between different canonical form methods is small, but varies significantly across different datasets. We also find that efficiency is the main difference between the methods.

Skeleton-based canonical forms for non-rigid 3D shape retrieval

“…Precision-recall curves for each method tested on the SHREC'15 canonical forms benchmark [3] using a view-based retrieval method [26]. Our method achieves the best precision for low and high recall values, falling below least-squares MDS for mid-range recall values.…”

“…In Section 4.1, we present our retrieval performance on the SHREC'15 canonical forms benchmark [3]. Secondly, in Section 4.2, we present the results Two dimensional illustration of the assignment regions for bone assignment.…”

“…Our method was recently entered into the SHREC'15 canonical forms benchmark [3]. The purpose of this benchmark was to compare the effectiveness of different methods at producing canonical forms for 3D shape retrieval.…”

“…We compare our canonical forms against those submitted to the SHREC'15 canonical forms benchmark [3], as we have their implementations. We omitted the least-squares MDS B method, as it is simply the same as the least-squares method, but with an early termination from the MDS algorithm which has shown to decrease the quality of the canonical forms [3]. Figure 5 shows two example meshes and their associated canonical forms produced using each method.…”

“…Fewer shape details are distorted, because the method effectively produces a set of canonical angles at the articulated joints of the shape, and shape deformations are localised to these joint regions. This leads to an increased retrieval performance on a recent canonical forms benchmark [3].…”