Eye tracking

Eye tracking is the ground truth of attention research. Most of what designers know about on-page scanning — the F-pattern, banner blindness, how little of a page is actually read — comes from eye-tracking studies, notably the large corpora published by Nielsen Norman Group.

A classic study is expensive and slow: hardware, a lab or calibrated webcams, recruited participants, and weeks from brief to report — which is why it has mostly stayed inside agencies and research teams rather than everyday design workflows.

The decades of recordings those studies produced became training data. Academic datasets such as MIT300 and SALICON pair images with real human gaze, and neural saliency models learn from them to predict fixation density on new images — the approach known as predictive eye tracking. Heatpoints runs one of these models (UNISAL, ECCV 2020) on a screenshot of your page and returns the predicted map in seconds.

The two are complements, not rivals: prediction gives instant, repeatable answers for the first-seconds, free-viewing question — where will the eye land? — while a live study remains the tool for task-driven behavior, comprehension, and anything a model cannot infer from pixels.