Active vision simulator

Watch a five-minute saliency-biased Lévy-like scanpath unfold through nonuniform peripheral sampling and the mechanisms that stabilize vision across saccades.

Choose an image

wally.webpYour image stays in this browser; it is never uploaded.Calculating 42 feature maps…

Calculating saliency0:00.0 / 5:00.0preparing scanpath

Click to restart from a chosen fixation · drop an image here

Playback speed1×

0:005:00

Ex-Gaussian fixationSaccadeμ + τ = 285 ms expected dwell

Evidence-based percept illustrationThis is a teaching model, not a literal reconstruction of conscious experience.

Mechanism mixer

Build the percept

Teaching preset

Presets affect rendering; the seeded scanpath is preserved.

01 · Spatial asymmetry

02 · Saccades & stability

03 · Saliency & timing

Itti–Koch–Niebur (1998)42 feature maps: 6 intensity + 12 color + 24 orientation; conspicuity maps combined at scale 4.

Brockmann–Geisel Figure 6 fit (1999)Proposal density ρ(Φ) ∝ Φ^−2.23; simulation truncated to the reported 2°–28° sample span; uniform directions are a 2D implementation assumption.

Fixation durationex-Gaussian with requested μ = 160, σ = 50, τ = 125 ms; expected mean = 285 ms.

Saccade durationMean = clamp(21 + 2.2 × amplitude, 25, 80) ms; truncated Gaussian residual SD = 15% of mean (minimum 3 ms; simulator assumption); sampled range = 15–120 ms.

—fixations—sample mean—mean amplitude

Each target starts with the paper’s Figure 6 power-law amplitude fit and a uniform direction, then is weighted by Itti saliency and short-lived inhibition of return. This is a 2D radial adaptation of ρ(x|y) ∝ s(x)f(|x−y|); finite-field rejection and IOR are explicit additions to Brockmann and Geisel’s one-dimensional model. The empirical fit is used because their theoretical Cauchy scale was not fitted in visual degrees.

Simulator constants, not Brockmann–Geisel estimates: up to 256 valid target proposals, saliency floor 0.045, IOR floor 0.03, 3.2° IOR width, 700 ms decay, and 3.2 s retention.

The 42-map feature architecture follows Itti et al.; orientation filtering uses a phase-invariant 9 × 9 quadrature adaptation (period 7 px, σ = 7/3) because the 1998 paper does not fix those kernel details.

04 · Display & field

Overlay full 5 min scanpathShows every fixation and saccade in other views; it is always visible in the 5 min scanpath view. Overlay saliency mapA display aid; it does not change the cached target map. Show eccentricity rings2°, 5°, 10°, and 20°.

Visual field extent: 220° horizontal × 133° vertical. Eccentricity is measured from fixation in this angular coordinate system.

Simulated eye movements

Five-minute scanpath distributions

Each saccade duration is sampled around its amplitude-dependent mean (15% coefficient of variation; simulator assumption), then peak velocity is derived from the minimum-jerk trajectory. Intersaccadic intervals are measured from one saccade’s offset to the next onset.

Amplitude × peak velocity

Preparing scanpath…

Saccade amplitude

Empirical Lévy proposal, after saliency and field constraints

Intersaccadic interval

Offset-to-onset interval · ex-Gaussian fixation timing

Primary literature

References by phenomenon

Primary sources cover saliency, the Lévy-flight scanpath proposal, fixation timing, saccade kinematics, and each perceptual mechanism. Browser-specific numerical choices remain explicit approximations.