Geometry-Contrastive GAN for Facial Expression Transfer

TL;DR

This paper introduces GC-GAN, a novel method that uses geometry-contrastive learning to transfer facial expressions across subjects while preserving identity, even with significant facial differences.

Contribution

The paper presents a new geometry-contrastive GAN that effectively transfers facial expressions across subjects by embedding geometry into the latent space, handling misalignments and shape differences.

Findings

Effective transfer of facial expressions across diverse subjects.

Preserves identity while changing expressions.

Handles large facial shape differences.

Abstract

In this paper, we propose a Geometry-Contrastive Generative Adversarial Network (GC-GAN) for transferring continuous emotions across different subjects. Given an input face with certain emotion and a target facial expression from another subject, GC-GAN can generate an identity-preserving face with the target expression. Geometry information is introduced into cGANs as continuous conditions to guide the generation of facial expressions. In order to handle the misalignment across different subjects or emotions, contrastive learning is used to transform geometry manifold into an embedded semantic manifold of facial expressions. Therefore, the embedded geometry is injected into the latent space of GANs and control the emotion generation effectively. Experimental results demonstrate that our proposed method can be applied in facial expression transfer even there exist big differences in facial shapes and expressions between different subjects.