Meta Deformation Network: Meta Functionals for Shape Correspondence

TL;DR

The paper introduces Meta Deformation Network, a novel deep learning approach for 3D shape matching that uses a meta-functional architecture to improve deformation quality and speed.

Contribution

It proposes a new meta-functional neural network architecture for shape correspondence that dynamically generates parameters, enhancing efficiency and deformation accuracy.

Findings

Outperforms conventional decoder designs on MPI-FAUST Inter Challenge.

Achieves faster execution speeds without sacrificing deformation quality.

Demonstrates the effectiveness of meta-functionals in 3D shape matching.

Abstract

We present a new technique named "Meta Deformation Network" for 3D shape matching via deformation, in which a deep neural network maps a reference shape onto the parameters of a second neural network whose task is to give the correspondence between a learned template and query shape via deformation. We categorize the second neural network as a meta-function, or a function generated by another function, as its parameters are dynamically given by the first network on a per-input basis. This leads to a straightforward overall architecture and faster execution speeds, without loss in the quality of the deformation of the template. We show in our experiments that Meta Deformation Network leads to improvements on the MPI-FAUST Inter Challenge over designs that utilized a conventional decoder design that has non-dynamic parameters.