# Accurate Face Rig Approximation with Deep Differential Subspace Reconstruction

arXiv: https://arxiv.org/abs/2006.01746 · HTML: https://mlsanity.com/paper/2006.01746/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 2 Jun 2020.

## Built on by (most cited first, 8 of 8)

- [HeadNeRF: A Real-time NeRF-based Parametric Head Model](https://mlsanity.com/paper/2112.05637/) (2021)
- [Learning Skeletal Articulations with Neural Blend Shapes](https://mlsanity.com/paper/2105.02451/) (2021)
- [Neural Face Rigging for Animating and Retargeting Facial Meshes in the Wild](https://mlsanity.com/paper/2305.08296/) (2023)
- [Prior-Guided Multi-View 3D Head Reconstruction](https://mlsanity.com/paper/2107.04277/) (2021)
- [Versatile Face Animator: Driving Arbitrary 3D Facial Avatar in RGBD Space](https://mlsanity.com/paper/2308.06076/) (2023)
- [Rig Inversion by Training a Differentiable Rig Function](https://mlsanity.com/paper/2301.09567/) (2023)
- [High-fidelity Interpretable Inverse Rig: An Accurate and Sparse Solution Optimizing the Quartic Blendshape Model](https://mlsanity.com/paper/2302.04820/) (2023)
- [HeadRecon: High-Fidelity 3D Head Reconstruction from Monocular Video](https://mlsanity.com/paper/2312.08863/) (2023)

## Abstract (arXiv)

To be suitable for film-quality animation, rigs for character deformation must fulfill a broad set of requirements. They must be able to create highly stylized deformation, allow a wide variety of controls to permit artistic freedom, and accurately reflect the design intent. Facial deformation is especially challenging due to its nonlinearity with respect to the animation controls and its additional precision requirements, which often leads to highly complex face rigs that are not generalizable to other characters. This lack of generality creates a need for approximation methods that encode the deformation in simpler structures. We propose a rig approximation method that addresses these issues by learning localized shape information in differential coordinates and, separately, a subspace for mesh reconstruction. The use of differential coordinates produces a smooth distribution of errors in the resulting deformed surface, while the learned subspace provides constraints that reduce the low frequency error in the reconstruction. Our method can reconstruct both face and body deformations with high fidelity and does not require a set of well-posed animation examples, as we demonstrate with a variety of production characters.