Generalizing Eye Tracking With Bayesian Adversarial Learning

doi:10.1109/CVPR.2019.01218

Paper

Generalizing Eye Tracking With Bayesian Adversarial Learning

Published Jun 1, 2019 · Kang Wang, Rui Zhao, Hui Su

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

UNKNOWN SJR score

79

Citations

14

Influential Citations

Full text Semantic Scholar

Abstract

Existing appearance-based gaze estimation approaches with CNN have poor generalization performance. By systematically studying this issue, we identify three major factors: 1) appearance variations; 2) head pose variations and 3) over-fitting issue with point estimation. To improve the generalization performance, we propose to incorporate adversarial learning and Bayesian inference into a unified framework. In particular, we first add an adversarial component into traditional CNN-based gaze estimator so that we can learn features that are gaze-responsive but can generalize to appearance and pose variations. Next, we extend the point-estimation based deterministic model to a Bayesian framework so that gaze estimation can be performed using all parameters instead of only one set of parameters. Besides improved performance on several benchmark datasets, the proposed method also enables online adaptation of the model to new subjects/environments, demonstrating the potential usage for practical real-time eye tracking applications.

Highly Cited

Study Snapshot

Key takeawayIntegrating adversarial learning and Bayesian inference into a unified framework improves generalization performance in appearance-based gaze estimation, enabling online adaptation to new subjects and environments.

PopulationOlder adults (50-71 years)

Sample size24

MethodsObservational

OutcomesBody Mass Index projections

ResultsSocial networks mitigate obesity in older groups.

Sign up to use Study Snapshot

Consensus is limited without an account. Create an account or sign in to get more searches and use the Study Snapshot.

Create an account

Paper

Generalizing Eye Tracking With Bayesian Adversarial Learning

Sign up to use Study Snapshot

References

Neuro-Inspired Eye Tracking With Eye Movement Dynamics

Our method, incorporating eye movement dynamics from neurological studies, improves generalization capabilities in eye tracking across different subjects and environments.

Appearance-Based Gaze Estimation via Evaluation-Guided Asymmetric Regression

The Asymmetric Regression-Evaluation Network (ARE-Net) improves eye gaze estimation performance by using two-eye asymmetry and adaptively adjusting the strategy based on performance evaluation.

RT-GENE: Real-Time Eye Gaze Estimation in Natural Environments

Our new real-time algorithm using appearance-based deep convolutional neural networks improves eye gaze estimation accuracy in natural environments, even on lower resolution images.

CIRA: An Architecture for Building Configurable Immersive Smart-Rooms

The Cognitive Immersive Room Architecture (CIRA) enables multi-modal smart-rooms with spatial and contextual intelligence, supporting nested contexts, multiple languages, and various physical environments.

Deep Pictorial Gaze Estimation

Our deep neural network architecture improves gaze estimation accuracy and robustness to variations in gaze, head pose, and image quality compared to state-of-the-art methods.

Citations

Semi-Supervised Multitask Learning Using Gaze Focus for Gaze Estimation

This study proposes two multitask learning frameworks for estimating gaze point and direction, achieving accurate gaze tracking through facial images.

Causal Representation-Based Domain Generalization on Gaze Estimation

CauGE framework effectively generalizes gaze estimation across domains by extracting domain-invariant features, achieving state-of-the-art performance in the domain generalization on gaze estimation benchmark.

From Feature to Gaze: A Generalizable Replacement of Linear Layer for Gaze Estimation

The Analytical Gaze Generalization framework (AGG) improves gaze estimation accuracy in cross-domain settings without requiring target domain data, effectively alleviating overfitting problems.

Unsupervised data labeling and incremental cross-domain training for enhanced hybrid eye gaze estimation

Unsupervised data labeling and incremental cross-domain training improve hybrid eye gaze estimation models for Augmentative and Alternative Communication, reducing gaze error from 4.87o to 3.95o.

Improving Domain Generalization on Gaze Estimation via Branch-out Auxiliary Regularization

Branch-out Auxiliary Regularization (BAR) improves gaze estimation's generalization capabilities without requiring direct access to target domain data, enhancing performance in uncontrolled environments.

Gaze from Origin: Learning for Generalized Gaze Estimation by Embedding the Gaze Frontalization Process

The GFAL framework, which embeds gaze frontalization as an auxiliary task, achieves state-of-the-art performance in gaze estimation across unseen datasets, outperforming SOTA gaze unsupervised domain adaptation methods.

SLYKLatent, a Learning Framework for Facial Features Estimation

SLYKLatent improves gaze estimation accuracy by 8.7% on Gaze360, rivaling top MPIIFaceGaze results, and surpassing existing methods by significant margins, with potential for human-robot interaction.