The "Psychophysics" folder contains four .XLSX files, each corresponding to one of the psychophysical measurements of our study (data of five subjects), and twenty .MAT files for the condition of Experiment 1 with one single feature (20 subjects). Each. MAT file contains: Column 1: number of trials Column 2: condition (1 = Salient Pattern DX , 2 = Salient Pattern SX) Column 3: subject answer (1 = SX, 2 = DX) Column 4: subject preference (1 = Salient pattern, 0 = Non-salient pattern) Column 5: name of pattern1 showed Column 6: name of pattern2 showed Column 7: Reaction times (sec) The "Eye Movements" folder contains twenty-eight .EDF files (four files for seven participants). The EyeLink EDF* files contain many types of data, including eye movement events (BLINKS, FIXATIONS, SACCADES), messages (including the “default” ones by Eyelink, and the custom made ones, defining the single trial’s specific properties: this type of messages starts with “Trialinfo”), button presses, and eye-data samples. Here we have used the (calibrated) x and y position data of the right eye. Custom-made matlab scripts (available upon request to the authors) have been used to parse the data into trials, filter eye position and estimate eye velocity, select exploitable trials, detect and analyze saccades, etc. The array containing the relevant parameters for all the saccades analyzed in the manuscript (for each subject) is available in the mat file (of the type “SXX_TOT.mat”). The EDF file format is a highly compressed binary format, intended for use with SR Research EyeLink viewers and applications. EDF files are not readable, they need to be transformed into a readable format (typically ASCII). A tool to transform edf files into ascii is freely available on the SR Research Support online forum site: https://download.sr-support.com/dispdoc/page25.html. The subfolder "Psychophysics_control" contains seven .MAT files. The .MAT files of the type “SXX_psychophysics.mat” contain, for each participant XX, 400-raws (one per trial): Column 1: duration of presentation of the target and distractor compounds (26ms) Column 2: Label for response correct (1=saccade to the target; 0=saccade to the distractor) Column 3: size, in pixels, of each model-derived feature Column 4: position of the salient compound or target (1=Right; 0=Left) Column 5: SNR, or fraction of optimal features within the 10-features compound