Eye

The eye is a seeing organ. There are multiple models for the human eye. The eye has a lens and a retina. The retina is largely a flat but curved surface that has little depth. The retina is the image sensor of the eye.

Eyes connect to the brain using the optic nerve.

The eye sets many of the design targets for head-mounted displays in virtual reality (VR) and augmented reality (AR). In a 2021 review of AR and VR displays, Xiong, Hsiang, He, Zhan and Wu list field of view, eyebox, angular resolution, dynamic range and correct depth cues as the "demanding human vision requirements" that these displays have to satisfy. Some headsets also track the eyes directly. Konrad, Angelopoulos and Wetzstein note that many wearable displays already integrate eye tracking to support foveated rendering or other gaze-contingent display modes, and eye tracking in VR is used both for hands-free interaction and for research on visual attention and cognition.

Reviewed 27 September 2026. Checked every anatomy, eye-movement, display-requirement, ocular parallax, foveated rendering and eye-tracking claim against the cited papers, textbooks and Sony and Apple announcements, and corrected citation metadata. About review dates.

Anatomy and optics

The cornea is the outermost clear layer of the eye, in front of the anterior chamber, iris and pupil. According to StatPearls, it is responsible for approximately 65% to 75% of the refraction of light as it passes through the eye; it performs the initial refraction onto the lens, which further focuses the light onto the retina. The focusing power of the eye changes through accommodation: contraction of the ciliary muscle changes the shape of the lens so that the point of focus moves between near and distant objects.

The retina lines the back of the eye. It is approximately 0.5 mm thick, and the whole retina is a circular disc between 30 and 40 mm in diameter. The photoreceptors (rods and cones) lie in its outermost layer, against the pigment epithelium, so light has to pass through the full thickness of the retina before it reaches them. The ganglion cells are the output neurons of the retina, and their axons run to the brain in the optic nerve. The fovea, a blood-vessel-free spot at the center of the area called the macula, lies about 17 degrees (4.5 to 5 mm) from the optic disc.

Photoreceptors are not spread evenly across the retina. In a study of eight human retinas from seven donors aged 27 to 44, Curcio, Sloan, Kalina and Hendrickson counted an average of 4.6 million cones (range 4.08 to 5.29 million) and 92 million rods (range 77.9 to 107.3 million) per retina. Peak cone density in the fovea averaged 199,000 cones per square millimeter, with large differences between individuals (100,000 to 324,000), and cone density was an order of magnitude lower 1 mm away from the foveal center. The rod-free zone at the center of the fovea had an average horizontal diameter of 0.350 mm, or 1.25 degrees. Xiong et al. note that the eye has high visual acuity only in the central fovea region, the property on which foveated displays are based.

Schematic eye models

Vision scientists and display engineers describe the optics of the eye with simplified schematic eyes. Konrad, Angelopoulos and Wetzstein note that several schematic eye models have been proposed, "each listing slightly different values for the cardinal points". They name the Gullstrand number 1, the Gullstrand-Emsley and the Emsley reduced eyes as some of the most popular models, in decreasing order of complexity. For rendering, the two points that matter most are the front nodal point, which is the eye's center of projection, and the center of rotation. Citing measurements by Fry and Hill (1962), the authors give the center of rotation as 14.7536 mm behind the cornea on average for emmetropic subjects, and state that the distance between the center of rotation and the center of projection is 7 to 8 mm in all of the models they compared. Their own experiments used the relaxed Gullstrand-Emsley eye, with a separation of 7.6916 mm.

Visual performance and display requirements

Resolution

Xiong et al. write that 20/20 visual acuity "amounts to 1 arcmin angular resolution, or 60 pixels per degree (PPD)", and that this figure is "considered as a common goal for AR and VR displays". A 2025 study in Nature Communications by Maliha Ashraf and Rafał Mantiuk of the University of Cambridge and Alexandre Chapiro of Meta Reality Labs measured the limit directly on a sliding display that allowed continuous control of resolution. They found that the limit for foveal vision was higher than previously believed: 94 pixels per degree for achromatic (black and white) patterns, 89 ppd for red-green patterns and 53 ppd for yellow-violet patterns. They also measured the limit at 10 and 20 degrees from the fovea and observed a much larger drop for the color patterns than for the achromatic ones. The authors describe this limit as what is often called the retinal resolution, although the limiting factor "may not necessarily be attributed to the retina".

Acuity falls away from the center of gaze. In their 2012 foveated graphics study, Guenter, Finch, Drucker, Tan and Snyder modeled the minimum detectable angular size as increasing linearly with eccentricity, and their user study gave a slope of 1.32 to 1.65 arc minutes per degree of eccentricity. Xiong et al. state that the central fovea region, where visual acuity is high, accounts for about 10 degrees of the field of view.

Field of view

For reference, Xiong et al. give the horizontal field of view of human vision as up to 160 degrees for one eye and 120 degrees for the region seen by both eyes (binocular overlap).

Interpupillary distance

The distance between the centers of the two pupils, the interpupillary distance (IPD), sets the separation of the two images in a stereoscopic display. In a 2004 survey of the available anthropometric data, Neil Dodgson found that mean adult IPD is around 63 mm, that the vast majority of adults fall between 50 and 75 mm, that a range of 45 to 80 mm is likely to include almost all adults, and that the minimum for children down to five years old is around 40 mm. He also noted that mean IPD values quoted in the stereoscopic literature range from 58 mm to 70 mm. A larger eyebox lets a headset tolerate more of this variation, as well as movement of the headset during use.

Eye movements

The Purves Neuroscience textbook describes four basic types of eye movement: saccades, smooth pursuit movements, vergence movements and vestibulo-ocular movements.

Type	What it does	Notes from the textbook
Saccades	Rapid, ballistic movements that abruptly change the point of fixation	About 200 ms passes between the appearance of a target and the start of the movement; the movement itself lasts on the order of 15-100 ms, and a target that moves during it is missed
Smooth pursuit	Slower tracking movements that keep a moving stimulus on the fovea	Under voluntary control, but most people cannot make a smooth pursuit movement without a moving target
Vergence	Disconjugate movements that converge or diverge the lines of sight to align both foveas on objects at different distances	Convergence is part of the near reflex triad, with accommodation of the lens and constriction of the pupil
Vestibulo-ocular	Reflex movements that stabilize the eyes relative to the world by compensating for head movement	Driven by the semicircular canals of the vestibular system; the eyes move the same distance as the head but in the opposite direction

During a saccade, vision is briefly suppressed. Road to VR described saccades as quick eye movements that "take tens of milliseconds", during which saccadic suppression renders the viewer effectively blind until the eye reaches its new point of fixation. Eye movements also complicate eye tracking in VR: Clay, König and König point out that in a 3D environment, concepts such as fixations and saccades are not as clearly defined as when a subject looks at a static 2D screen, and that fixations must be separated from smooth pursuit and optokinetic nystagmus.

Accommodation and vergence

When a person looks at a near object, three responses happen together: the eyes converge, the ciliary muscle contracts to change the shape of the lens, and the pupil constricts. StatPearls calls this the accommodation reflex, also known as the accommodation-convergence reflex or the near reflex.

Hoffman, Girshick, Akeley and Banks note that the uncoupling of vergence and accommodation required by 3D displays frequently reduces the ability to fuse the binocular stimulus and causes discomfort and fatigue. Two separate images drive vergence to the depth of the virtual object, but the image itself sits at the fixed optical distance of the display, which leads to incorrect accommodation cues. Xiong et al. call the mismatch the vergence-accommodation conflict (VAC) and note that it may be more serious in AR than in VR, because the virtual image is superimposed directly on a real world that has correct depth cues. Hoffman and colleagues built a 3D display that presents correct or nearly correct focus cues and used it to test the effect. With correct focus cues, the time needed to identify a stereoscopic stimulus fell, stereoacuity in a time-limited task rose, distortions in perceived depth were reduced, and viewer fatigue and discomfort were reduced. Clay, König and König describe the same issue for VR headsets, where the scene is presented on a single plane at a fixed distance and the focus of the lens therefore carries no depth information. Proposed solutions include multifocal and varifocal displays, holographic displays and integral imaging displays, which produce an accommodation cue, and Maxwellian-view displays, which remove the accommodation cue altogether; Xiong et al. regard multifocal and varifocal displays as a relatively practical solution for VR headsets.

Ocular parallax

The optical center of the eye is not the same as the rotational center, giving rise to ocular parallax, which has been accounted for by some researchers who have built an eye tracking system to automatically change the rendering of a VR scene slightly according to someone's gaze.

When the eye rotates to look at a different part of a scene, its center of projection moves around the center of rotation, creating small, depth-dependent shifts of the image on the retina. The cue was first described by David Brewster in 1845. Konrad, Angelopoulos and Wetzstein at Stanford University implemented "ocular parallax rendering" on an HTC Vive Pro fitted with a Pupil Labs eye tracker. The renderer computes the position of each eye's nodal point from the tracked gaze and adjusts the view frustum accordingly, so it adds no computational cost to a headset that already tracks the eyes. In their user studies, detection thresholds for the effect were almost an order of magnitude lower than visual acuity at the same point in the peripheral visual field, and the authors concluded that it is "clearly visible in most VR applications". Ocular parallax rendering improved the ordering of relative depth and the impression of realistic depth, but did not necessarily improve absolute distance estimates. The authors also note that gaze-contingent distortion in head-mounted displays, attributed in part to ocular parallax by earlier researchers, is commonly known as pupil swim.

The eye in VR and AR systems

Foveated rendering

Because acuity is high only near the fovea, a system that knows where the user is looking can render the periphery at lower detail. Guenter et al. rendered several image layers at different sampling rates around the gaze point, which they reported accelerated graphics computation by a factor of 5 to 6 on a desktop HD display and reduced the number of shaded pixels by a factor of 10 to 15. They predicted that the technique would let rendering cost grow roughly linearly rather than quadratically as display field of view increases. Xiong et al. describe the related idea of a foveated display, in which a high-resolution image is projected only to the fovea while the peripheral image stays at low resolution. Sony's January 2022 announcement of the PlayStation VR2 listed eye tracking and foveated rendering among the headset's features.

Eye tracking for input and identification

Eye movements are also used as an input method. Sony stated that PS VR2 detects the motion of the eyes, "so a simple look in a specific direction can create an additional input for the game character". In the Apple Vision Pro, users browse apps by looking at them and tapping their fingers to select; Apple states that its eye tracking system uses high-speed cameras and a ring of LEDs that project invisible light patterns onto the user's eyes. The same headset uses the iris for Optic ID, which analyzes the iris under invisible LED light and compares it with enrolled data to unlock the device.

Eye data raises privacy questions. A survey by Bozkir and colleagues on eye-tracked VR states that eye-tracking data "reveals users' privacy-sensitive attributes when combined with the information about the presented stimulus", and reviews privacy-preserving methods and eye-based authentication for VR. See Privacy in virtual and augmented reality.

Saccadic redirection

Saccadic suppression has been used for redirected walking. Sun and colleagues presented a system at SIGGRAPH 2018 that uses a head- and eye-tracking VR headset to detect saccadic suppression and redirect users during the resulting temporary blindness. The authors reported that saccades can significantly increase the rotation gains during redirection without introducing visual distortions or simulator sickness, allowing large virtual spaces to be explored in small physical rooms. Road to VR reported that the system rotates the scene by a few degrees per saccade while the user walks.

See also

• Human visual system
• Eye tracking
• Foveated rendering
• Fovea
• Retina
• Accommodation
• Vergence-accommodation conflict
• Interpupillary distance
• Eye box
• Eye relief
• Pixels per degree
• Depth perception