Knowing When to Trust a Prior: Reliability-Gated Cue Fusion for Video Gaze Prediction

Key points

• 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.1
• 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.1
• Gated fusion is significantly positive on film, sports and web video, whereas unconditional fusion is harmful on sports and null on web.1
• 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.1

Sources (1)

• [1]Knowing When to Trust a Prior: Reliability-Gated Cue Fusion for Video Gaze Prediction
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 04:46 PM

  We propose FocusGate, a gated ensemble of gaze-free priors whose members may abstain.

  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.

Extractive summary: sentences quoted from the sources.