Adaptation Speed for Exposure Control in Virtual Reality

Claus B. Madsen

and Johan Winther Kristensen

Computer Graphics Group, Aalborg University, Aalborg, Denmark

Keywords:

Virtual Reality, Computer Graphics, Exposure Control, Dynamic Range.

Abstract:

We address the topic of real-time, view-dependent exposure control in Virtual Reality (VR). For VR to real-

istically recreate the dynamic range of luminance levels in natural scenes, it is necessary to address exposure

control. In this paper we investigate user preference regarding the temporal aspects of exposure adaptation.

We design and implement a VR experience that enables users to individually tune how fast they prefer the

adaptive exposure control to respond to luminance changes. Our experiments show that 60% of users feel the

adaptation signiﬁcantly improves the experience. Approximately half the of users prefer a fast adaptation over

about 1-2 seconds, and the other half prefer a more gradual adaptation over about 10 seconds.

1 INTRODUCTION

Virtual Reality (VR) technology has witnessed un-

precedented growth in recent years, revolutionizing

various ﬁelds including gaming, education, health-

care, and industry. The fundamental appeal of VR

lies in its ability to transport users to synthetic en-

vironments, creating a sense of presence and immer-

sion. However, achieving a seamless and comfortable

user experience in VR necessitates overcoming sev-

eral technical challenges. One critical aspect is the

management of exposure levels within the virtual en-

vironment.

Exposure control encompasses the manipulation

of visual parameters such as brightness, contrast, and

dynamic range to ensure that the visual content pre-

sented to the user aligns with their physiological and

perceptual capabilities. In VR, accurate exposure

control is paramount for achieving a realistic impres-

sion of an environment in terms of its dynamic range

of luminance values.

This paper investigates how VR users evaluate as-

pects of exposure adaptation in VR. Speciﬁcally, we

design and implement an experiment which enables

users to tune their personal preference regarding how

quick and responsive the exposure adaptation should

be in VR. The core contribution of this research lies

in demonstrating that VR users prefer exposure adap-

tation and that they feel it increases the realism of the

experience, in addition to giving speciﬁc guidelines

for the temporal aspects of such adaptation.

https://orcid.org/0000-0003-0762-3713

The paper is organized as follows. In section 2

we expand on the background for this work and de-

scribe related work. Section 3 then brieﬂy describe

the speciﬁc goals of this research and the approach

taken. In section 4 we go through how we designed

and implemented the VR experience used in the pre-

sented experiments. The design of the experiment is

presented in section 5, followed by a presentatio of

experimental results in section 6. Finally, section 7

offers a conclusion.

2 BACKGROUND AND RELATED

WORK

The human eye can handle a vast range of luminance

levels, where luminance here is the photometric con-

cept of candela per square meter. In radiometry the

corresponding concept would be radiance, but in this

paper we shall stick to photometric terminology. With

the rods and the cons in the retina, the human eye can

adapt to widely varying light conditions, from the dim

glow of starlight to the intense glow sunlight So, the

real physical world has a huge dynamic range in terms

of luminance levels, (Reinhard et al., 2010).

Unfortunately, cameras do not at all support the

same dynamic range. Cameras have to adjust the ex-

posure in order to ﬁnd a sensitivity that is suitable for

a given scene; and this exposure can be a compromise,

for example when taking pictures indoor the windows

may end up over-exposed. And probably everyone

have experience how the camera on their smartphone

Madsen, C. and Kristensen, J.

Adaptation Speed for Exposure Control in Virtual Reality.

DOI: 10.5220/0012450300003660

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2024) - Volume 1: GRAPP, HUCAPP

and IVAPP, pages 307-312

ISBN: 978-989-758-679-8; ISSN: 2184-4321

307

will automatically adjust the exposure to ensure a de-

cent image given the luminance levels of the scene.

Humans also adapt their own sensitivity to luminance

levels when, e.g., walking from an indoor to an out-

door environment.

Similarly, various types of displays, TVs, cinema

screens, computer monitors, VR headsets, etc., do

not support close to real world dynamic range, either.

They simply cannot recreate luminance levels of the

same magnitude as found in, say, and outdoor day-

light scene.

Tone mapping and exposure adaptation are two

terms that relate to the challenges of having to display

high dynamic range content on low dynamic range

displays. The goal of tone mapping is to, as closely

as possible, on a given display, recreate the visual im-

pression the viewer would have if they observed the

real scene (Haines and Hoffman, 2018). In this paper

we focus solely on exposure adaptation, i.e., adjust-

ing the light sensitivity in response to changing lumi-

nance levels.

In ﬁlm, exposure control can be used as a means

for supporting viewer understanding of lighting con-

ditions in a scene and how they change over time,

(Bordwell et al., 2008). And in VR, we deﬁnitely also

need to understand the pros and cons of exppsure con-

trol and tonemapping, if we want VR users to move

around naturalistic looking scenes, and getting close

to realistic perception of changes in luminance levels.

A method for view-dependent tone-mapping for

in the case of 360-degree video was presented in

(Najaf-Zadeh et al., 2017). This work concluded

that users preferred a view-dependent over a global,

ﬁxed exposure for the whole sequence. Other re-

search has focused on the challenges with adapting

existing 2D single-image tone mapping approaches

to view-dependent 360-degree and VR context. The

challenges are related to achieving global consistency,

while retaining the beneﬁts of the view-dependent

mapping, (Goud

e et al., 2019; Goud

e et al., 2020;

Melo et al., 2018).

It would seem that there has been quite some re-

search, although not a massive amount, into various

technical aspects of employing exposure control and

tone mapping in VR contexts. But, we have not really

been able to ﬁnd studies of whether users ﬁnd such

adaptive techniques appealing, or conducive to an en-

hanced sense of realism. Hence, the purpose of this

paper was to investigate some of these aspects.

3 OVERVIEW OF APPROACH

The main research question behind this work was: If

users experience a VR environment, where the ren-

dering continuously performs view-dependent expo-

sure adaptation such that center-view luminance lev-

els are properly exposed, at what rate do these users

then prefer the adaptation to happen? Should it hap-

pen quickly? Or in a more slow and subtle way? We

were curious to ﬁnd out, if we might experimentally

determine a user concensus regarding this.

To experiment with this we designed a VR experi-

ence, where the user should be able to freely adjust the

speed with which the exposure adaptation happens.

The VR scenario should be familiar to the user, and

it should entail some realistic luminance ranges. Our

idea was then to subject a number of users to this envi-

ronment and study if there was any systematic trends

regarding which adaptation speed they preferred.

4 IMPLEMENTATION

For the purposes of conducting user experiments into

the aspects of real-time dynamic tone mapping we

designed and implemented a VR experience. The

VR experience is single static space with no mov-

ing objects, but with illumination elements that of-

fer various levels of luminance values, from a dark

ﬂoor in shadow under a table, to ceiling mounted light

sources, and a projector projecting a photo onto a

wall. We used the Unity game engine for the imple-

mentation of the environment.

4.1 Scenario Design

We decided to let the VR environment for the tests be

a recreation of the actual physical space where the ex-

periments would take place. For two reasons. Firstly,

because it was convenient to have a real physical ver-

sion of what the VR environment should like in terms

of dimensions, materials and illumination. And sec-

ondly, there is plenty of literature supporting the fact

that sense of presence and ability to sense for exam-

ple distances in VR is heightened by using transition

environments, where the experience in VR is a recre-

ation of the physical space the user ”comes from”

when entering the VR, (Steinicke et al., 2009; Okeda

et al., 2018; Soret et al., 2021).

The chosen space was an 8x8x4 meter room with

no windows, so there is no natural daylight coming

into the room. All elements in the VR scene had sim-

ilar sizes, materials and placement as their real world

GRAPP 2024 - 19th International Conference on Computer Graphics Theory and Applications

308

Figure 1: View of the VR environment developed for the

experiment. The enviroment is a virtual recreation of the

physical 8x8x4 meter room actually used for the experi-

ments. Towards the right in this view it is possible to see

how a project is projecting an color photograph onto a wall.

counterparts. Figure 1 shows a view of the room from

one corner.

In the virtual scene there were three types light

sources: 1) ceiling lights, 2) a table lamp aimed at a

table, and 3) a projector light aimed at a wall.

The ceiling and the table lights were each imple-

mented in Unity as two lights: One to light up the

scene, and one with a very small range to light up ﬁx-

ture/light origin without spilling light into the scene.

For the ceiling and table lights, the two light sources

pointed in opposite directions: one pointed down-

wards to spread light into the scene, and one pointed

upwards to light up the ﬁxture. Ceiling lights were set

to 650 lumen for the upwards light with a small range

(1 meter as the holes in the ﬁxture should allow for

light on the ceiling) while the downwards light was

set to 1900 lumen with a range of 10 meters. The

table light upwards light was set to 1557 lumen at a

range of 0.06 meters and a downwards light of 2850

lumen and a range of 10 meters.

The projector in the scenario was implemented in

Unity as projector light at 40000 lumen and set up to

use a High Dynamic Range image as a decal, while a

point light at 295 lumen was placed at the front of the

projector to light up the front and glass.

All lumen values were chosen based on subjective

experience of the environment, not on actual mea-

sured values from real world lights. The lighting in

all the scenes was baked to allow more natural light-

ing with global illumination.

4.2 Rendering and Exposure Control

The scenario was built using the High Deﬁnition Ren-

der Pipeline (HDRP) in Unity, as the HDRP has built-

in functions for controlling exposure based on the

light visible in the virtual camera. The HDRP has

several modes of controlling the exposure control,

Figure 2: Top: the user is looking towards the dark ﬂoor

in the VR room, and, with the center of the ﬁeld of view

being so dark, the dynamic adaptation adjust the exposure

resulting parts of the wall becoming over-exposed and sat-

urated. Bottom: the user is looking towards a projection on

the wall, and the dynamic exposure control has adapted to

the much higher luminance levels in the center of the view.

where the Automatic mode was used for this study. In

Automatic mode, exposure control is done based on

the within-frame content of the scene. In Automatic

mode we opted to utilize the option of letting the ex-

posure control be center-weighted such that whatever

is in the center of the ﬁeld-of-view at any given time is

weighted higher when computing the proper exposure

adjustment.

Unity generates the environment for VR by calcu-

lating each eye as its own camera, which means that

in some cases one eye might have different light ex-

posure to the other eye. To combat this we selected a

softness of 5 so that pixels on the edge of the mask had

less inﬂuence. This way, having a light at the edge of

the camera would have less inﬂuence on the exposure

which is similar to the eye’s adaptation. The adapta-

tion speed for the exposure was set to be between 0.01

seconds and 15 seconds, as any speed higher or lower

had no discernible differences in adjustment. The test

participants only controlled the speed from dark to

light, with the speed from light to dark being set as

one third of their chosen value. This was to recreate

the standard human adaptation as our eyes more eas-

ily adapt to light than to dark environments.

Adaptation Speed for Exposure Control in Virtual Reality

309

4.3 Interaction and Adjustment Options

For the purposes of our experiments we wanted three

types of interaction with the VR environment: 1) be-

ing able to freely look and move around the scenario,

2) being able to turn lights in the scene on and off to

experiment with how the exposure control looked and

felt, and 3) being able to adjust the speed with which

the exposure control adapts to changes in luminance

levels.

With regards to moving and looking around we

opted to base all navigation on the tracking from the

VR headset, so the participants had 6 degree of free-

dom movement, but no ability to do for example tele-

porting. So, all visual exploration of the scene was

based on physically moving and looking around.

In terms of allowing participants to switch lights

on or off, this was implemented as button presses on

the controllers: using one face button on the right

hand controller and two face buttons on the left hand

controller, such that in total three lights could be tog-

gled (desk lamp, ceiling lights, projector). Changing

the status of a light was actually implemented as a

load of a different scene, in order to have the light-

ing solution baked for increase performance and vi-

sual realism. So, we had scenes baked corresponding

to all combinations of lights being on or off, respec-

tively.

As mentioned in section 4 participants could ad-

just adaptation speed in the interval from 0.01 sec-

onds (essentially instantaneous) to 15 seconds (almost

too slow to be discernible). The current adaptation

speed was visualized to test participants via an opaque

sphere inside a large semi-transparent sphere locked

to the position of their right controller. The inner

sphere changed size depending on the current adap-

tation speed, with a higher speed meaning a larger

sphere, and the outer sphere representing the fastest

possible value, Figure 3. This way participants could

see if they were close to the maximum or minimum

speed available.

A tutorial scene was available to users with no

prior VR experience, allowing them to get familiar

with the VR controllers. This scene used the same

setup as the other scenes, although the projector light

was turned off. Here the user could press the buttons

to squish and stretch a sphere or move a cube up and

down using the joystick. A total of ﬁve participants

tried the training scenario, each of which had never

tried VR or only a short time before.

Figure 3: A view of the test participants adaptation speed

as symbolised with an opaque sphere inside a transparent

sphere. The opaque sphere would change size depending

on the chosen speed, with a high speed resulting in a large

sphere and vice versa. The maximum speed would set the

opaque sphere at the same size as the transparent sphere.

5 EXPERIMENT DESIGN

The aim of the experiment was two-fold: 1) what

adaptation speed does each test person prefer if there

is dynamically adapting tone mapping, and 2) does

the test person prefer dynamic adaptation, or not. The

experiment was run with 21 participants of varying

age (22 to 32, mean 27) and varying levels of VR ex-

perience (novice to experienced). The Meta Quest 2

VR headset was used for all experiments.

The room used for all experiments was identical

to the scene that test participants were going to expe-

rience in VR, in terms of size, materials and illumi-

nation, etc.. So, once test participants put on the VR

headset the whole VR scene and its appearance would

be completely familiar to test participants.

Each test consisted of 2 phases. Phase 1 started

with a short introduction to the experiment, and gath-

ering the age and previous VR experience of the par-

ticipant. Each participant was then offered to try a

training scene to familiarize themselves with using

the controllers. Five participants with little to no prior

VR experience opted to try the training scene.

Once in the actual Phase 1 VR test scenario, test

participants were instructed how to move around, how

to turn individual lights in the scene on and off, and

how to adjust the exposure adaptation speed, in the

range described in section 4. Participants were not

told about adaptation speed values or had access to

GRAPP 2024 - 19th International Conference on Computer Graphics Theory and Applications

310

any numerical read-out. When adjusting the adapta-

tion speed they could just see the solid sphere inside

the transparent sphere visually indicating the current

value relative to the maximum, as described in sec-

tion 4.3. Participants were then given the time they

wanted to move and look around, explore, and exper-

iment with setting the adaptation speed to whatever

value they preferred. Once a participant had settled

on a preferred adaption speed they were allowed to

take off the VR headset, and the chosen adaptation

speed would be logged.

After this, the participant would be instructed of

the objective of Phase 2. Here the participants would

be exposed once more to the exact same VR scene, but

this time they could not adjust adaptation speed, they

could only toggle dynamic adaptation on/off, where

on would be using the adaptation speed the participant

had elected as preferred in Phase 1, and off would

be no dynamic adaptation at all, only a static generic

exposure value chosen by experimenters to visually

mimic an impression of the general luminance level

of the real room. Participants were then given the

time they wanted to move and look around, explore,

turn lights on and off, and toggle dynamic adaptation.

Once satisﬁed, participants would be allowed to take

off the VR headset, asked if they preferred dynamic

adaptation or not, and their answer to this would be

logged. Subsequently, the participant be asked to rate

the how much the adaptation improved they experi-

ence, and how natural the adaptation felt. In both

cases using a Likert 1 - 7 scale where 1 would be not

at all, and 7 would be signiﬁcantly. And this would

conclude the test session for the participant.

6 RESULTS AND DISCUSSION

Participants tested the adaptation speed in various

ways. Most of the participants looked toward the pro-

jector in the virtual space to test the speed when go-

ing from dark to bright, as it was the brightest source

in the scene. Some participants turned to look at the

ﬂoor to get the dark adaptation, as the ﬂoor was dark.

Other participants turned the lights on and off instead,

although not so many utilized this option.

Figure 4 shows a histogram over what adaptation

speeds test participants preferred in Phase 1 of the

experiment. Participants tend to fall in two groups.

Either they prefer a slow and subtle adaption, or the

prefer a faster, clearly visible adaptation. The pre-

ferred values do not conform to a normal distribution

in a Shapiro-Wilk test (p = 0.07). More test persons

would be needed to establish whether this parameter

is actually a bi-modal distribution.

Figure 4: Histogram over preferred adaptation speeds

logged during Phase 1 of the experiment.

Figure 5: Histogram over test participant responses to the

question ”To what extent do you feel the adaptation im-

proved the visual experience?”, using a 1-7 Likert scale,

with 1 being not at all, and 7 being signiﬁcantly.

The clear results from Phase 2 was that 17 out of

21 participants preferred dynamic adaptation over no

adaptation. Three preferred without, and one was un-

decided. This particular participant set an adaptation

speed at its lowest possible value, making the adap-

tation happen over several seconds, and therefore the

participant had a hard time telling when the adapta-

tion was on or off. The participant commented on this

but chose to continue with the very low speed. Fig-

ure 5 shows how participants scored to what extent

the adaptation improved the experience, with approx.

60% scoring signiﬁcantly (6 and 7).

After the experiment, participants were also asked

to rate how certain they felt in their choice of pre-

ferred adaptation speed, using a 1-7 Likert scale.

Roughly one third scored their certainty around 3 to 4,

and the remaining participants scored their certainty

high (5, 6, or 7). There was no statistically signiﬁcant

correlation between time spent in the test environ-

Adaptation Speed for Exposure Control in Virtual Reality

311

ment, and self-reported certainty with preferred adap-

tation speed. There was also no statistically signiﬁ-

cant correlation between previous VR experience and

self-reported certainty in preferred adaptation speed.

7 CONCLUSIONS

We designed a VR experience aimed at testing aspects

of user preference regarding aspects of dynamic tone

mapping in VR applications. The VR experience was

a virtual recreation of the physical room used for user

experiments. The VR experience allowed users to in-

dividually set how fast they preferred the dynamic lu-

minance adaptation be, ie., how quickly they wanted

adaptation to respond to drastic changed in luminance

values between different parts of the VR scene, for ex-

ample between looking at a dark ﬂoor and then shift-

ing your viewing direction towards a light source or a

bright wall.

The experiment clearly showed that test partici-

pants prefer dynamic adaptation, and that they feel

this adaptation signiﬁcantly improves the experience.

Out of 21 test participants, 17 preferred the version of

the VR experience that had dynamic luminance adap-

tation over the version that did not. And 60% reported

that it signiﬁcantly improved the experience.

The experiment seemed to indicate that partici-

pants fall in two groups in terms of preferred adapta-

tion speed, i.e., how quickly they want the adaptation

to respond to luminance changes. About half of the

participants preferred a slow, relatively subtle adapta-

tion, whereas the other half preferred a faster, clearly

perceptible, and more immediate adaptation response.

For future work it would be extremely interest-

ing to investigate whether does dynamic adaptation

somehow tricks test participants into believing that

the dynamic range of luminance in the scene is ac-

tually higher than what the VR headset can recreate.

For example, is a light source in a VR scene perceived

brighter when experienced with dynamic luminance

adaptation than without.

ACKNOWLEDGEMENTS

This research was partially funded by the CityVR

project, funded by the Department of Architecture,

Design and Media Technology, Aalborg University,

and partially by the BUMUS project, funded by the

Danish Ministry of Food, Agriculture and Fishery.

This funding is gratefully acknowledged.

REFERENCES

Bordwell, D., Thompson, K., and Smith, J. (2008). Film

art: An introduction, volume 8. McGraw-Hill New

York.

Goud

e, I., Cozot, R., and Banterle, F. (2019). Hmd-tmo:

a tone mapping operator for 360 hdr images visu-

alization for head mounted displays. In Computer

Graphics International Conference, pages 216–227.

Springer.

Goud

e, I., Lacoche, J., and Cozot, R. (2020). Tone map-

ping high dynamic 3d scenes with global lightness co-

herency. Computers & Graphics, 91:243–251. ID:

271576.

Haines, E. and Hoffman, N. (2018). Real-time rendering.

Crc Press.

Melo, M., Bouatouch, K., Bessa, M., Coelho, H., Cozot, R.,

and Chalmers, A. (2018). Tone mapping hdr panora-

mas for viewing in head mounted displays. In Pro-

ceedings of the 13th International Joint Conference

on Computer Vision, Imaging and Computer Graphics

Theory and Applications, volume 1, pages 232–239.

Scitepress.

Najaf-Zadeh, H., Budagavi, M., and Faramarzi, E. (Sep

2017). Vr+hdr: A system for view-dependent render-

ing of hdr video in virtual reality. pages 1032–1036.

IEEE.

Okeda, S., Takehara, H., Kawai, N., Sakata, N., Sato, T.,

Tanaka, T., and Kiyokawa, K. (2018). Toward more

believable vr by smooth transition between real and

virtual environments via omnidirectional video. In

2018 IEEE International Symposium on Mixed and

Augmented Reality Adjunct (ISMAR-Adjunct), pages

222–225. ID: 1.

Reinhard, E., Heidrich, W., Debevec, P., Pattanaik, S.,

Ward, G., and Myszkowski, K. (2010). High dynamic

range imaging: acquisition, display, and image-based

lighting. Morgan Kaufmann.

Soret, R., Montes-Solano, A.-M., Manzini, C.,

Peysakhovich, V., and Fabre, E. F. (2021). Pushing

open the door to reality: On facilitating the tran-

sitions from virtual to real environments. Applied

Ergonomics, 97:103535. ID: 271441.

Steinicke, F., Bruder, G., Hinrichs, K., Steed, A., and Ger-

lach, A. L. (2009). Does a gradual transition to the

virtual world increase presence? In 2009 IEEE Vir-

tual Reality Conference, pages 203–210. IEEE.

GRAPP 2024 - 19th International Conference on Computer Graphics Theory and Applications

312