A Generalized Label Shift Perspective for Cross-Domain Gaze Estimation

2025-05-19

AI Summary

This paper addresses the degradation in generalization performance caused by domain shift in cross-domain gaze estimation. Departing from conventional feature-invariance assumptions, it pioneers a generalized label shift (GLS) perspective to model joint distribution discrepancies between source and target domains. The proposed correction framework employs truncated Gaussian importance reweighting and introduces a probabilistic-aware conditional operator discrepancy metric to achieve conditional invariance learning. Unlike prior methods, it is agnostic to specific backbone architectures. Extensive experiments on multiple standard cross-domain gaze estimation (CDGE) benchmarks demonstrate substantial improvements in cross-domain accuracy and robustness, validating its model-agnosticism and strong generalization capability. The core contribution lies in the first integration of GLS theory into gaze estimation, establishing a novel paradigm for cross-domain gaze modeling that transcends traditional domain adaptation assumptions.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Intelligent Robots: State Estimation

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Aiming to generalize the well-trained gaze estimation model to new target domains, Cross-domain Gaze Estimation (CDGE) is developed for real-world application scenarios. Existing CDGE methods typically extract the domain-invariant features to mitigate domain shift in feature space, which is proved insufficient by Generalized Label Shift (GLS) theory. In this paper, we introduce a novel GLS perspective to CDGE and modelize the cross-domain problem by label and conditional shift problem. A GLS correction framework is presented and a feasible realization is proposed, in which a importance reweighting strategy based on truncated Gaussian distribution is introduced to overcome the continuity challenges in label shift correction. To embed the reweighted source distribution to conditional invariant learning, we further derive a probability-aware estimation of conditional operator discrepancy. Extensive experiments on standard CDGE tasks with different backbone models validate the superior generalization capability across domain and applicability on various models of proposed method.

Problem

• Generalizing gaze estimation models to new domains
• Addressing label and conditional shift in cross-domain gaze estimation
• Overcoming continuity challenges in label shift correction

Innovation

• Introduces Generalized Label Shift perspective
• Uses truncated Gaussian reweighting strategy
• Derives probability-aware conditional discrepancy estimation