Depth of field

Depth of field (DOF) is the range of distances in a scene over which objects appear acceptably sharp in an image formed by a lens, whether that lens belongs to a camera or to the human eye. Points nearer or farther than that range are imaged as blur spots larger than a chosen maximum allowable blur, called the circle of confusion.

In virtual reality (VR) and augmented reality (AR), depth of field matters for two linked reasons. A conventional head-mounted display (HMD) places all of its imagery at a single optical distance, so the focus cues the eye receives, accommodation and retinal blur, specify the depth of the display rather than the depths in the depicted scene. This is closely tied to the vergence-accommodation conflict. Researchers have responded with rendered depth-of-field blur driven by eye tracking, with optics that move or multiply the focal plane, and with displays whose images stay sharp at any focus.

Reviewed 4 October 2026. Bibliographic details and abstracts of all cited papers (Crossref, OpenAlex, PubMed, Europe PMC), quotes and figures from the Stanford, Nikon, Epic Games, Meta, UploadVR, Road to VR, VentureBeat and GitHub sources. About review dates.

Definition

In camera optics, a point that is not at the focused distance projects onto the sensor as a blurred spot. A feature counts as out of focus once the width of its blurred image on the sensor exceeds a chosen maximum allowable circle of confusion. Depth of field is the span of object-space distances inside that limit; the matching span on the image side of the lens is the depth of focus. Nikon's microscopy reference makes the same distinction, describing depth of field as the distance from the nearest object plane in focus to the farthest plane simultaneously in focus, and noting that depth of field refers to object space while depth of focus refers to image space, although the two terms are often used interchangeably.

Three variables govern it. Closing the aperture (a larger f-number) increases depth of field. Depth of field varies inversely with the square of focal length, so long lenses have very shallow depth of field, and it varies with the square of the subject distance, so distant scenes are sharp over a much larger range than close-ups.

The eye's depth of field

The human eye is also a lens system with a finite aperture, so it has a depth of field. Vision science uses both terms for the eye's tolerance to defocus, which is usually expressed in diopters. A 2006 review by Bin Wang and Kenneth J. Ciuffreda in Survey of Ophthalmology defines the eye's depth-of-focus as "the perceptual tolerance of the human eye to retinal defocus."

Measured values depend on pupil size and on the blur criterion. In a 2014 study in Optometry and Vision Science, Bernal-Molina and colleagues measured the depth of field of seven eyes from young subjects, using an objectionable-blur criterion. As accommodative demand rose from 0 to 6 diopters, mean pupil diameter fell from 5.70 to 4.62 mm and the eye's depth of field changed from 0.85 ± 0.26 D to 1.07 ± 0.19 D. The authors concluded that, when focusing a near target, the visual system uses this tolerance to change its refractive state less than would be needed for a perfectly focused image, and that accommodation aims for a retinal image that is "good enough."

Depth of field in head-mounted displays

In a real scene, the eye focuses at one distance and objects at other distances fall into progressively stronger retinal blur. In a stereoscopic display the light comes from one surface. Hoffman, Girshick, Akeley and Banks (2008) pointed out that the focus cues, accommodation and the blur in the retinal image, then specify the depth of the display rather than the depths in the depicted scene. Using a display that presented correct or nearly correct focus cues, they found that viewers identified stereoscopic stimuli faster, had better stereoacuity in a time-limited task, perceived depth with less distortion, and reported less fatigue and discomfort.

Most consumer headsets behave this way. In a 2018 UploadVR report on Facebook's varifocal research, Ian Hamilton wrote that "Nearly all consumer HMDs present a single fixed focus," that most lock the optical focus to something around two meters, and that most people report some blur when their eyes focus on a near virtual object.

Blur is a depth cue in its own right. Held, Cooper, O'Brien and Banks (2010) showed that the pattern of blur in an image, combined with relative depth cues, indicates the apparent scale of a scene, and that applying blur to a sharply rendered image changes the apparent distance and scale of its contents. Zannoli, Love, Narain and Banks (2016) found that observers could not reliably judge which of two surfaces was nearer at an occlusion boundary when the blur was rendered onto a single display plane, but did much better when the same scenes were shown on multiple planes at different focal distances so that the blur was produced by the eye's own optics; chromatic aberration in the eye supplied part of the useful information.

Rendering depth of field

Origins in computer graphics

Early computer graphics used a pinhole camera model in which everything is in focus. Michael Potmesil and Indranil Chakravarty extended it in 1981 with a camera model that approximates the effects of a lens and aperture, so that synthetic images could have a depth of field and be focused on an arbitrary plane. In 1984 Robert L. Cook, Thomas Porter and Loren Carpenter introduced distributed ray tracing, which spreads ray directions according to the function being sampled and so produces depth of field, motion blur, penumbras, translucency and fuzzy reflections, with depth-of-field calculations integrated into the visible-surface computation.

Depth-of-field effects in VR applications

In Unreal Engine, depth of field is a post-process effect. Epic Games' VR best-practice guidance tells developers to "Avoid post process effects such as Depth of Field and Motion Blur to prevent user discomfort." The Unreal Engine 4.27 version of the page also lists r.DepthOfFieldQuality=0 among the console settings from Epic's Showdown VR demo that it offers as a starting point for VR projects.

Gaze-contingent depth of field

Gaze-contingent depth of field uses an eye tracker to find the point the user is looking at, sets the rendered focal distance to the depth of that point, and blurs the rest of the scene accordingly. Sébastien Hillaire, Anatole Lécuyer, Rémi Cozot and Géry Casiez presented an early version at the 2008 IEEE Virtual Reality conference: they retrieved the user's focus point in a 3D virtual environment with an eye-tracking system and used it to drive a depth-of-field blur that "simulates the fact that humans perceive sharp objects only within some range of distances around the focal distance." In their experiment participants generally preferred the effects when they were dynamically adapted to the focus point. The same authors published "Depth-of-Field Blur Effects for First-Person Navigation in Virtual Environments" in IEEE Computer Graphics and Applications in November 2008. Radosław Mantiuk, Bartosz Bazyluk and Anna Tomaszewska described gaze-dependent depth-of-field rendering for virtual environments in 2011.

Later studies measured what the technique does for perception and comfort:

Study	Venue	Display	Summary
Mauderer, Conte, Nacenta and Vishwanath (2014)	ACM CHI 2014	Regular display	Gaze-contingent DOF increased subjective realism and perceived depth and contributed to judgments of depth order and distance between objects, but with limited accuracy.
Vinnikov and Allison (2014)	ACM ETRA 2014	Stereoscopic display	Built a gaze-contingent DOF system to test the hypothesis that simulated focal blur set by instantaneous fixation can compensate for the fixed accommodation cue of a stereoscopic display, and studied the user experience in realistic scenes.
Duchowski, House, Gestring, Wang, Krejtz, Krejtz, Mantiuk and Bazyluk (2014)	ACM Symposium on Applied Perception 2014	Stereoscopic display	Gaze-contingent DOF significantly reduced subjective visual discomfort, yet participants disliked it, which the authors attributed to eye-tracker inaccuracy and the blur simulation's noticeable lag.
Weier, Roth, Hinkenjann and Slusallek (2018)	ACM Transactions on Applied Perception	Head-mounted display	A foveated rendering system with an integrated gaze-contingent DoF filter was rated almost on par with full rendering while shading more than 69% fewer samples.

ChromaBlur

Standard depth-of-field rendering mimics a high-quality camera. Steven A. Cholewiak, Gordon D. Love, Pratul P. Srinivasan, Ren Ng and Martin S. Banks argued in 2017 that rendering for immersive displays should instead model the optics of the human eye, in particular its chromatic aberration, which colors the blur differently for objects nearer or farther than the current focus. In their experiments the resulting method, ChromaBlur, drove accommodation "quite effectively," whereas conventionally rendered blur did not drive it at all, and viewers judged ChromaBlur images as having greater depth and realism. The authors proposed combining it with focus-adjustable lenses and gaze tracking in HMDs to reduce the effects of vergence-accommodation conflict.

DeepFocus

A varifocal display moves its single focal plane to wherever the user looks, so the fixated object is sharp, but by itself it does not blur the rest of the scene. Facebook Reality Labs (now Meta's Reality Labs) built DeepFocus to fill that gap for its Half Dome prototype, a varifocal headset with a 140 degree field of view whose displays physically move to shift focus. The December 2018 announcement, now hosted on Meta's blog, said that "while varifocal VR headsets can deliver a crisp image anywhere the viewer looks, DeepFocus allows us to render the rest of the scene just the way it looks in the real world: naturally blurry," and described it as the first system to produce this gaze-contingent defocus effect realistically and in real time.

DeepFocus is a convolutional neural network by Lei Xiao, Anton Kaplanyan, Alexander Fix, Matthew Chapman and Douglas Lanman, published in ACM Transactions on Graphics for SIGGRAPH Asia 2018. Its paper notes that varifocal, multifocal and light field HMDs all extend depth of focus but rely on expensive rendering to reproduce accurate defocus blur; DeepFocus synthesizes defocus blur, focal stacks, multilayer decompositions and multiview imagery from ordinary RGB-D images. Meta said the network was trained on 196,000 images from a random scene generator and initially ran at 1080p, and that VR-quality output in real time was demonstrated on a four-GPU machine. The code and dataset were released on GitHub under the non-commercial CC-BY-NC 4.0 license; the repository was archived as read-only on 31 October 2023. Road to VR reported that the technique was intended to mitigate the eyestrain associated with vergence-accommodation conflict.

Display approaches

Rendered blur on a fixed-focus display does not change the optical distance of the image itself. Several display designs change the focal distance of the displayed light, or remove the eye's dependence on it:

• Varifocal displays track gaze and move one focal plane to the fixated depth, pairing with rendered blur for the rest of the scene.
• Multifocal (multiplane) displays present images at several focal distances at once. Kurt Akeley, Simon Watt, Ahna Girshick and Martin Banks built a 2004 stereo prototype with three image planes at different physical distances that gave near-correct focus cues without tracking eye position. Narain and colleagues (2015) used a model of defocus in the viewer's eye to optimize the images on each plane so that retinal images at different accommodation states match those of the original scene, including scenes with occlusions.
• Light field displays are the third architecture the DeepFocus paper names alongside varifocal and multifocal designs; the paper notes that all three extend depth of focus but need computationally expensive rendering to reproduce accurate defocus blur.
• Accommodation-invariant displays do the opposite of reproducing focus cues. Robert Konrad, Nitish Padmanaban, Keenan Molner, Emily A. Cooper and Gordon Wetzstein (2017) engineered a near-eye display whose stimulus does not change with the eye's accommodation state, so that accommodation can be driven by stereoscopic cues instead and the vergence-accommodation mismatch is significantly reduced. They validated the principle with a prototype that measured users' accommodation while they viewed different display modes.
• Maxwellian displays, also called accommodation-free displays, focus the image through the pupil of the eye and keep it sharp regardless of the viewer's focal distance, but typically have a small eye box. In the head-mounted design described by Shrestha and colleagues (2019), narrow collimated pixel beams are focused through the pupil, so each spot, and therefore the raster image, is perceived as in focus regardless of accommodation. The authors note that a perfectly aligned Maxwellian display allows only half a pupil diameter of lateral movement before vignetting occurs.

A 2019 state-of-the-art report in Computer Graphics Forum by George Alex Koulieris, Kaan Akşit, Michael Stengel, Rafał Mantiuk, Katerina Mania and Christian Richardt surveys the vision science and display engineering behind these near-eye designs.

Capture and production

Depth of field also shapes how XR content is captured. A light field camera samples the 4D light field rather than a single focused image: the hand-held plenoptic camera described by Ren Ng, Marc Levoy and colleagues in a 2005 Stanford technical report places a microlens array between the main lens and the sensor, which lets sharp photographs focused at different depths be computed after a single exposure and extends depth of field without reducing the aperture.

In virtual production on LED volumes, a physical camera films a wall that already shows rendered imagery. Epic Games' documentation notes that this creates two depths of field, one rendered by the virtual camera on the wall and one from the physical lens, giving a result blurrier than expected. Its experimental in-camera VFX depth-of-field compensation feature adjusts the circle of confusion for each pixel from the camera's distance to the wall and the pixel's depth, following the real camera's focus distance, aperture and focal length.

See also

• Depth perception
• Focal surface display
• Stereoscopy
• Lytro