Saliency map
also: attention map, saliency model
A saliency map is a grayscale prediction of where human eyes are most likely to look in an image, brightest where attention concentrates. It is produced by a saliency model trained on eye-tracking data, without any real viewer present.
Saliency modeling comes from computer vision and neuroscience: given a picture, predict the fixation density a crowd of viewers would produce in the first seconds of free viewing. The output is a heatmap — hot where the eye lands, cold where it skips.
Modern models are neural networks trained on datasets like SALICON and MIT300, where thousands of images were shown to people whose gaze was recorded. UNISAL (ECCV 2020), the model behind Heatpoints, scores AUC-J 0.872 / NSS 2.322 on MIT300 at a fraction of the size of competing networks.
For web pages, a saliency map answers a practical question before you have any traffic: does the eye land on your headline and CTA, or on a decorative image? It is a prediction, not a recording — pair it with real analytics to close the loop.
See it on your own page
Free score /100 · no signup · results in seconds