3D-Aware Face Editing via Warping-Guided Latent Direction Learning

Yuhao Cheng, Zhuo Chen, Xingyu Ren, Wenhan Zhu, Zhengqin Xu, Di Xu, Changpeng Yang, Yichao Yan,
MoE Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University
Huawei Cloud Computing Technologies Co., Ltd
CVPR 2024
Paper Code (Coming Soon)
Teaser

An example of our warping-guided 3D-aware face editing method. Our method supports users to edit 3D faces in an intuitive way that drags points from multiple perspectives. Moreover, our method can achieve disentangled editing for shape, expression, and view, while maintaining 3D consistency.

Abstract

3D facial editing, a longstanding task in computer vision with broad applications, is expected to fast and intuitively manipulate any face from arbitrary viewpoints following the user's will. Existing works have limitations in terms of intuitiveness, generalization, and efficiency.

To overcome these challenges, we propose FaceEdit3D, which allows users to directly manipulate 3D points to edit a 3D face, achieving natural and rapid face editing. After one or several points are manipulated by users, we propose the tri-plane warping to directly deform the view-independent 3D representation. To address the problem of distortion caused by tri-plane warping, we train a warp-aware encoder to project the warped face onto a standardized latent space. In this space, we further propose directional latent editing to mitigate the identity bias caused by the encoder and realize the disentangled editing of various attributes. Extensive experiments show that our method achieves superior results with rich facial details and nice identity preservation. Our approach also supports general applications like multi-attribute continuous editing and cat/car editing.

Pipeline

Pipeline
we propose FaceEdit3D, an intuitive method to edit the 3D facial shape and expression from any perspective. Our approach involves a tri-plane warping to ensure the inherent 3D-consistent editing. To mitigate facial distortions led by the warping, we train a warp-aware encoder to project the warped face into standardized distribution and further explore the hierarchical mechanism in latent space to achieve disentangled editing. Extensive experiments demonstrate the effectiveness and efficiency of our method. The additional applications also show the generalization and potential of our method across different applications. To sum up, our method provides a brand new way to manipulate the 3D representation, opening up new avenues for rapid and convenient real-image editing.

BibTeX

@InProceedings{Cheng_2024_CVPR,
  author    = {Yuhao, Cheng and Zhuo, Chen and Xingyu, Ren and Wenhan, Zhu and Zhengqin, Xu and Di, Xu and Changpeng, Yang and Yichao, Yan},
  title     = {3D-Aware Face Editing via Warping-Guided Latent Direction Learning},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2024},
}