Makeup Prior Models for 3D Facial Makeup Estimation and Applications

TL;DR

This work introduces two types of makeup prior models to extend existing 3D face prior models: PCA-based and StyleGAN2-based priors and designs a makeup consistency module to address the challenges that previous methods faced in robustly estimating makeup.

Abstract

In this work, we introduce two types of makeup prior models to extend existing 3D face prior models: PCA-based and StyleGAN2-based priors. The PCA-based prior model is a linear model that is easy to construct and is computationally efficient. However, it retains only low-frequency information. Conversely, the StyleGAN2-based model can represent high-frequency information with relatively higher computational cost than the PCA-based model. Although there is a trade-off between the two models, both are applicable to 3D facial makeup estimation and related applications. By leveraging makeup prior models and designing a makeup consistency module, we effectively address the challenges that previous methods faced in robustly estimating makeup, particularly in the context of handling self-occluded faces. In experiments, we demonstrate that our approach reduces computational costs by several orders of magnitude, achieving speeds up to 180 times faster. In addition, by improving the accuracy of the estimated makeup, we confirm that our methods are highly advantageous for various 3D facial makeup applications such as 3D makeup face reconstruction, user-friendly makeup editing, makeup transfer, and interpolation.

Figures (20)

• Figure 1: Example of 3D facial makeup estimation and applications using makeup prior models. Top left: The effectiveness of our prior models (PCA and StyleGAN2) for estimating 3D facial makeup layers. Compared to the existing method MakeupExtract (Yang-Ext and Yang-Res), our method robustly estimates makeup layers, especially in the case of self-occluded faces. Bottom left: The result of 3D face reconstruction using makeup prior models. Our method accurately recovers the makeup of 3D faces and it can be compatible with the existing 3D face reconstruction framework DECA. Right: 3D makeup interpolation and transfer applications using the PCA-based prior model. Note that the StyleGAN2-based prior model has equivalent functionality.
• Figure 2: Examples of makeup sampled from PCA and StyleGAN2 prior models. The top row demonstrates visual composites $\mathbf{M}_{v}$ from the combination of makeup bases $\mathbf{M}_{b}$ and alpha matte $\mathbf{M}_{a}$. The bottom row shows variations produced by manipulating the coefficient in the prior models.
• Figure 3: Overview of the makeup estimation network architecture. The network is composed of three modules: The Reconstruction module is pre-trained for 3D face reconstruction; The makeup estimation module employs ${E}_{make}$ to infer the makeup coefficient $\mathbf{\xi}$ and generates associated makeup textures; The makeup consistency module enhances the effectiveness of makeup estimation.
• Figure 4: Comparison with 3D facial makeup estimation methods. The results demonstrate the robustness of our methods (PCA and StyleGAN2) in terms of stability and accuracy in estimating makeup, outperforming both Yang-Ext and Yang-Res MakeupExtract, which show limitations in handling self-occluded faces.
• Figure 5: Comparison with 3D face reconstruction methods using makeup prior models. Our methods successfully reconstruct facial makeup. Specifically, the PCA model is capable of broadly recovering makeup colors, while the StyleGAN2 model achieves precise replication of complex makeup features, such as blush and gradient eyeshadow.