Facial Expression Retargeting From Human to Avatar Made Easy

Area of Science:

• Computer Graphics
• Animation
• Machine Learning

Background:

• Facial expression retargeting is crucial for virtual character animation.
• Traditional methods (markers, blendshapes) are labor-intensive and require expert 3D modeling skills.

Purpose of the Study:

• To develop an efficient and user-friendly method for cross-domain facial expression transfer.
• To overcome limitations of traditional 3D modeling approaches in expression retargeting.

Main Methods:

• Utilized variational autoencoders to create low-dimensional latent spaces for human and avatar facial expressions.
• Established correspondences between latent spaces using geometric and perceptual constraints.
• Developed a user-friendly system for automatic triplet generation and annotation to capture perceptual preferences.

Main Results:

• Successfully trained a network for expression domain translation from human to avatar.
• Demonstrated high-quality facial expression retargeting results.
• User studies confirmed the method's effectiveness for nonprofessional users.

Conclusions:

• The proposed method offers a novel solution for facial expression retargeting via nonlinear embedding and domain translation.
• Achieves high-quality results with reduced time and effort compared to traditional techniques.
• Empowers nonprofessional users to perform complex expression transfer tasks effectively.