Search papers, labs, and topics across Lattice.
This paper introduces Factor-Informed Uncertainty Distillation (FIUD), a novel teacher-student framework designed to enhance gaze estimation in unconstrained environments by aligning uncertainty predictions with interpretable image-quality factors. By employing a gradient-boosting teacher to predict expected gaze error based on factors like illumination and eye visibility, the method enables a neural student to effectively distill these signals into a lightweight uncertainty head through curriculum learning. The results show significant improvements in uncertainty and error rank correlation across multiple datasets, particularly in challenging settings, outperforming both deterministic and sampling-based approaches.
Aligning uncertainty with interpretable image-quality factors leads to significant improvements in gaze estimation accuracy, especially in unconstrained environments.
Deep gaze estimation works well in controlled capture but degrades in unconstrained settings, where systems must reject unreliable predictions. Single-pass uncertainty (e.g., heteroscedastic regression) infers uncertainty from pixels without explicit input-validity cues, while sampling based methods are often too costly for real time use. We propose Factor-Informed Uncertainty Distillation (FIUD), a teacher-student framework that aligns uncertainty with interpretable image-quality failure modes. A gradient-boosting teacher predicts expected gaze error from factors such as illumination, sharpness, eye visibility and symmetry; a neural student distills these signals via curriculum learning and ranking supervision into a lightweight single-pass uncertainty head. Across ETH-XGaze, Gaze360, and MPIIFaceGaze (> 300k samples), FIUD improves uncertainty, error rank correlation and selective prediction versus deterministic and sampling-based baselines, with the largest gains in unconstrained settings.
