Learning to predict where humans look

TLDR

Understanding human gaze is crucial for graphics, design, and HCI, yet existing saliency models rely on bottom‑up cues and ignore top‑down semantics, leading to poor alignment with real eye movements when eye tracking is unavailable. The study aims to learn a saliency model that incorporates low, middle, and high‑level image features by training on eye‑tracking data from 15 viewers across 1003 images. The authors collected eye‑tracking data from 15 participants on 1003 images and used this dataset to train and test a saliency model based on multi‑level image features. The resulting eye‑tracking database, comprising 1003 images and 15 viewers, is released publicly alongside the paper.

Abstract

For many applications in graphics, design, and human computer interaction, it is essential to understand where humans look in a scene. Where eye tracking devices are not a viable option, models of saliency can be used to predict fixation locations. Most saliency approaches are based on bottom-up computation that does not consider top-down image semantics and often does not match actual eye movements. To address this problem, we collected eye tracking data of 15 viewers on 1003 images and use this database as training and testing examples to learn a model of saliency based on low, middle and high-level image features. This large database of eye tracking data is publicly available with this paper.