An active foveated gaze prediction algorithm based on a Bayesian ideal observer

•A model to predict human fixations in a visual search task for objects in real-world scenes, tested and evaluated on COCO-search18 dataset.•A theory-based model of active fixation prediction in a foveated visual system which chooses fixations to maximise information gain at each step.•A model based on a region-proposal network which uses eccentricity-based max pooling to represent the loss of detail in peripheral vision.•The presented model does not need to be trained on human eye movement datasets and so can generalize to any set of images and targets through transfer learning.