FocusGate improves video gaze prediction by trusting reliable priors
Knowing When to Trust a Prior: Reliability-Gated Cue Fusion for Video Gaze Prediction
Computer Vision and Pattern Recognition
Summary
Video gaze prediction models usually rely on training data with recorded eye movements, but they can miss useful clues from other sources. The authors introduce FocusGate, which decides when to trust different gaze-free signals based on frame quality. This method improves gaze prediction in various types of videos, like movies and sports, without much added delay. FocusGate can also outperform some existing gaze prediction models when used alone.
What this means in practice
- •For video analytics developers: Improve gaze tracking accuracy in video-based applications by selectively trusting gaze-free cues when they are reliable.
- •For human-computer interaction engineers: Enhance interactive systems that respond to user attention by integrating reliable gaze priors with trained gaze models on diverse video content.
Authors
Lichen Zhu, Yueqian Lin, Yiheng Wang, Hai "Helen" Li, Yiran Chen
Abstract
Video gaze prediction is led by gaze-trained models, yet gaze-free priors carry signal those models have not absorbed, if one knows when to trust them. We propose FocusGate, a gated ensemble of gaze-free priors whose members may abstain. A per-frame gate reads three shape statistics of a defocus map and selects the frames on which the estimator is above chance on average, so rejected frames reduce to the base exactly, while midrank normalisation lets an all-zero prior abstain at zero parameters. Gated fusion is significantly positive on film, sports and web video, whereas unconditional fusion is harmful on sports and null on web. Added to four supervised predictors, the NTIRE 2026 champion among them, FocusGate improves all sixteen model-domain cells in shuffled AUC, fifteen significantly, one domain pre-registered and scored once, while adding only 1% to the champion's latency. Alone, it surpasses TASED-Net and UNISAL in shuffled AUC on film with a 16-frame causal mean.