SHREC'17: Deformable Shape Retrieval with Missing Parts

The final report paper is available for download here.

Matlab code for the baseline approach is available for download here.

Ground-truth classification files are now available for download here.


A challenging setting of 3D shape retrieval occurs when the shapes to be retrieved are allowed to undergo non-rigid deformations, which may range from simple changes in pose to more complicated transformations such as intra-class variation and topological noise. Even more challenging is the setting in which the objects in question have missing parts in addition to the aforementioned deformations. In this scenario, the query object is a subset of the full model and the task is to retrieve similar objects with different partiality patterns (or lack thereof), additionally undergoing a non-rigid transformation; vice versa, the query object might be a full model, while the shape repository may contain deformed partial views of objects of the same class.

With this SHREC'17 track we propose a benchmark to evaluate the performance of partial shape retrieval methods, under the concurrent presence of:

  • non-rigid deformations
  • different amounts and types of partiality (including simulated rangemaps)
  • topological changes induced by mesh gluing in areas of contact


Main contact: Emanuele RodolĂ  (USI Lugano) < emanuele dot rodola at usi dot ch >

Co-organizers: Or Litany (Tel Aviv University), Michael Bronstein (USI Lugano), Alex Bronstein (Technion).