Siggraph Asia 2022: Rig Inversion by Training a Differentiable Rig Function
This paper was accepted for publication and presentation at Siggraph Asia on 6 December 2022.
Authors: Mathieu Marquis Bolduc and Hau Nghiep Phan.
Rig Inversion by Training a Differentiable Rig Function
Download the full research paper. (1 MB PDF)
Rig inversion is a mathematical approach that allows animators to remap an existing mesh animation onto an animation rig. This allows animators to tweak and fix up existing mesh animations, which would be difficult or impossible otherwise.
The difficulty with rig inversion is finding the rig parameter vector that best approximates a given input mesh.
In this paper, we propose to solve this problem by first obtaining a differentiable rig function. We do this by training a multi-layer perceptron to approximate the rig function. This differentiable rig function can then be used to train a deep-learning model of rig inversion.