The dataset contains 25 non-rigidly deformed versions of a “null” shape (i.e. a shape in a neutral pose with no topological changes) undergoing different topological changes. The shapes are partly taken from the KIDS dataset but any self-intersections are removed, creating topological changes on different parts of the shape.

A subset of 15 shapes is given as training data. For each shape in the training set the correspondence to the null shape is given. Additionally, ground-truth correspondence mapping each point to its left-right symmetric counterpart is also provided. Note that, as a result of the merging, some points may not have a symmetric match.

Due to the changes, not all vertices in the modified shapes have a reliable ground-truth match onto the null shape. A map indicating these vertices is included in the data set (see inset figure - yellow indicates points with no reliable matches).

The whole data set exists in two different resolutions, ~10k vertices and ~60k-80k vertices.



In the contest the shapes 1-15 were given as training data and 16-25 were used for testing. The evaluation code allows you to put your method in comparison to the methods that participated in the contest. Please cite the following paper if you are using this data for your research

author={ Z. L\"ahner and E. Rodol\`a and M. M. Bronstein and D. Cremers and O. Burghard and L. Cosmo and A. Dieckmann and R. Klein and Y. Sahillioglu },
title={ SHREC’16: Matching of Deformable Shapes with Topological Noise },
booktitle={ Proc. of Eurographics Workshop on 3D Object Retrieval (3DOR) },
year={ 2016 }