Learning Gaze-aware Compositional GAN

TL;DR

This paper introduces a Gaze-aware Compositional GAN that generates annotated facial images from limited labeled data, improving gaze estimation models through data augmentation and enabling facial editing and gaze redirection.

Contribution

The novel Gaze-aware Compositional GAN effectively leverages limited labeled and unlabeled data to generate annotated facial images for gaze estimation and related applications.

Findings

Effective in generating within-domain augmentations for ETH-XGaze.

Successful cross-domain augmentation for CelebAMask-HQ.

Enables facial editing and gaze redirection applications.

Abstract

Gaze-annotated facial data is crucial for training deep neural networks (DNNs) for gaze estimation. However, obtaining these data is labor-intensive and requires specialized equipment due to the challenge of accurately annotating the gaze direction of a subject. In this work, we present a generative framework to create annotated gaze data by leveraging the benefits of labeled and unlabeled data sources. We propose a Gaze-aware Compositional GAN that learns to generate annotated facial images from a limited labeled dataset. Then we transfer this model to an unlabeled data domain to take advantage of the diversity it provides. Experiments demonstrate our approach's effectiveness in generating within-domain image augmentations in the ETH-XGaze dataset and cross-domain augmentations in the CelebAMask-HQ dataset domain for gaze estimation DNN training. We also show additional applications of our work, which include facial image editing and gaze redirection.