Welcome to Remote Calibrator! This package contains several useful tools to calibrate and track for the remote psychophysics experiments, e.g., crowd-sourced through Amazon Mechanical Turk.
Please visit https://calibrator.app for the demo. More information can be found at https://easyeyes.app/remote-calibrator/.
If you use it, please cite us.
Jiang, P., Burchell, A., & Pelli, D. G. (2021, December). EasyEyes aids online testing by emulating a wireless keyboard and tracking viewing distance. In PERCEPTION (Vol. 50, No. 1_ SUPPL, pp. 7-7). 1 OLIVERS YARD, 55 CITY ROAD, LONDON EC1Y 1SP, ENGLAND: SAGE PUBLICATIONS LTD.
@inproceedings{jiang2021easyeyes,
title={EasyEyes aids online testing by emulating a wireless keyboard and tracking viewing distance},
author={Jiang, Peiling and Burchell, Augustin and Pelli, Denis G},
booktitle={PERCEPTION},
volume={50},
number={1\_ SUPPL},
pages={7--7},
year={2021},
organization={SAGE PUBLICATIONS LTD 1 OLIVERS YARD, 55 CITY ROAD, LONDON EC1Y 1SP, ENGLAND}
}
To use Remote Calibrator, you can add the following line into the <head>
of your HTML file.
<script src="https://cdn.jsdelivr.net/npm/remote-calibrator@latest"></script>
Or use package management tools, e.g., NPM.
npm i remote-calibrator
And import the package to your project
import RemoteCalibrator from 'remote-calibrator'
Either way, you will then be able to use functions listed below under RemoteCalibrator
. For example,
RemoteCalibrator.init({ id: 'session_022' })
RemoteCalibrator.measureDistance({}, data => {
console.log(`The viewing distance is ${data.value}cm.`)
})
Task | Functions |
---|---|
๐ฌ Initialize | init() (always required) |
๐ฑ Panel | async panel() removePanel() resetPanel() |
๐ฅ๏ธ Screen | Display Pixel Dimensions screenSize() |
๐ Viewing Distance | measureDistance() |
๐ Distance Tracking | (viewing distance and near point) trackDistance() async getDistanceNow() Nudger Lifecycle Others |
๐ Gaze | trackGaze() async getGazeNow() calibrateGaze() nudgeGaze() Lifecycle Others |
๐ป Environment | System and Browser Environment |
๐ Customization | backgroundColor() videoOpacity() showCancelButton() |
๐ Other Functions | performance() checkInitialized() getFullscreen() newLanguage() |
๐ฃ Getters | Experiment Performance Environment i18n All Data Others |
Arguments in square brackets are optional, e.g. init([options, [callback]])
means both options
configuration and the callback
function are optional, but you have to put options
, e.g., {}
, if you want to call the callback function. The default values of options
are listed in each section with explanation.
Please be advised that some of the package documentation here may be outdated, although this is not a frequent occurrence. We are currently in the process of developing a typed interface for the package, which will provide you with up-to-date information on the APIs as you use them.
.init([options, [callback]])
Initialize RemoteCalibrator. Must be called before any other functions and can only run once. Return this
.
Pass { value, timestamp, date }
(equivalent to RemoteCalibrator.id
) to callback.
/* [options] Default value */
{
/**
* The id of the session, a string
* Will be attached to all the data from calibration
* A random one will be generated if no value is passed into the function
*/
id: /* Randomized value */,
/**
* Set the language, e.g., 'en-US', 'zh-CN'
* If set to 'AUTO' (default), the calibrator will try to follow the browser settings
* A full list of supported languages can be found at
* https://docs.google.com/spreadsheets/d/1UFfNikfLuo8bSromE34uWDuJrMPFiJG3VpoQKdCGkII/edit#gid=0
*/
language: 'AUTO',
/**
* Enter full screen if set to true
* Will be ignored if already in full screen mode
*/
fullscreen: false,
}
The callback function will be called after the initialization. Like many other functions below, one argument (an object) will be passed into it then. Please see the example.
function initializationFinished(data) {
// data: { value, timestamp, date }
console.log(`RemoteCalibrator was initialized at ${data.timestamp}.`)
}
let options = { id: 'session_022' }
RemoteCalibrator.init(options, initializationFinished)
If you do not want to change anything in default options, simply use an empty object like this:
RemoteCalibrator.init({}, initializationFinished)
The data
passed into the callback function is an object with three fields: date
, timestamp
and value
(the id). The date
is a JavaScript Date
object with all the information from the year to the millisecond, at the time when Remote Calibrator is initialized. You can find how to get these information here. The timestamp
(as all the timestamp
fields mentioned below) is the result of performance.now()
called at the same time of the respective execution. Learn more about its difference compared to Date
here.
While date
is only provided here, the absolute time of the following calibration actions can be calculated based on the initialization date, initialization timestamp, and the calibration timestamp.
/* async */ .panel(tasks, parentQuery, [options, [callback, [resolveOnFinish]]])
.panel()
is a powerful tool to help you set up a graphical user interface for participants to go through step-by-step and calibrate or set up tracking. It is highly customizable: tasks, task order, title, description, and "Done" button can all be customized. It is appended to the parent HTML node as set by parentQuery
, e.g., if the parent node has id main-area
, put #main-area
as the parentQuery
. Can only run once. Return a JavaScript Promise
that will resolve the resolveOnFinish
once the "Done" button is pressed.
tasks
is an array of tasks which can be a string or an object. Valid names are screenSize
, measureDistance
, trackDistance
, trackGaze
.
[
'screenSize',
{
name: 'trackGaze',
options: { framerate: 60 }, // Same as setting the options for .trackGaze()
callbackTrack: gotGaze, // Same as setting the callback for .trackGaze()
},
// Tracking viewing distance accepts two callbacks just like .trackDistance()
{
name: 'trackDistance',
callbackStatic: gotBlindSpotResult,
callbackTrack: gotTrackResult,
color: '#3490de',
},
]
You can customize the panel element with the following options.
/* [options] Default value */
{
headline: `Please press a button, to calibrate.`,
description: ``,
color: '#3490de', // Main color of the panel
showNextButton: false, // If set to false, will execute the callback right after the last calibration task
nextHeadline: `Thanks for calibrating. Hit the button to continue.`,
nextDescription: ``,
nextButton: `Done`,
debug: false,
i18n: true,
}
callback
will be called when the next-step button is clicked, which is disabled until all the calibration steps are finished.
If you don't want to use the default panel and want to integrate the process into your experiment, you can also call each calibration function individually. Please see the instructions below.
You can also use .removePanel()
to remove the panel element after the calibration is done, e.g., in the callback of .panel()
function.
.resetPanel([tasks, [parentQuery, [options, [callback, [resolveOnFinish]]]]])
helps you reset the panel to its initial state, i.e. none of the buttons were pressed. You may also pass in new tasks, options, etc., to refresh the element. However, this will not end the current trackings.
Get the display width and height in pixels. You can use .displayWidthPx
.displayHeightPx
.windowWidthPx
.windowHeightPx
getters. For example, RemoteCalibrator.windowWidthPx.value
will give you the inner width of the current browser window.
.screenSize([options, [callback]])
Get the screen width and height in centimeters. Like many other calibration functions, this function will pop an overlay interface for participants to use. The callback function will be called after the calibration process (the participant presses SPACE in this case).
Pass { value: { screenWidthCm, screenHeightCm, screenDiagonalCm, screenDiagonalIn, screenPpi, screenPhysicalPpi }, timestamp }
to callback. screenPpi
relates to the pixel data used in JavaScript, and screenPhysicalPpi
is the actual PPI of some Retina displays.
/* [options] Default value */
{
// Enter fullscreen if set to true
// Will be ignored if already in fullscreen mode
fullscreen: false,
// How many times the participant needs to calibrate
repeatTesting: 1,
// The length of decimal place of the returned value
decimalPlace: 1,
// Default virtual object used to measure screen size, can be `card`, `usba`, or `usbc`
defaultObject: "card",
// Headline on the calibration page (Support HTML)
headline: "๐ฅ๏ธ Screen Size Calibration",
// Description and instruction (Support HTML)
description: "...",
check: false, // Ask participants to use a ruler to measure against RC results
}
Before measuring or tracking viewing distance, calibration of the screen size is required to get the accurate value.
.measureDistance([options, [callback]])
Pop an interface for participants to calibrate the viewing distance at the moment using Blind Spot Test.
Pass { value, timestamp, method, raw }
(equivalent to RemoteCalibrator.viewingDistanceCm
) to callback.
/* [options] Default value */
{
fullscreen: false,
// How many times each of the eye will be used to test
// By default, right eye 2 times, then left eye 2 times
repeatTesting: 1,
sparkle: true, // Whether the moving dot flickers or not
decimalPlace: 1,
control: true, // Set to true to use arrow keys to control the position of the red dot
headline: "๐ Measure Viewing Distance",
description: "...",
check: false,
checkCallback: false,
showCancelButton: true,
}
.trackDistance([options, [callbackStatic, [callbackTrack]]])
Measure the viewing distance and then predict the real-time distance based on the change of the inter-pupillary distance, measured by face landmarks. callbackStatic
is called after getting the blind spot result and callbackTrack
is called every time a new result from estimation is derived.
Pass { value: { viewingDistanceCm, nearPointCm: { x, y }, latencyMs }, timestamp, method }
to callback.
method
can be either "BlindSpot"
(for measures from blind spot tests) or "FaceMesh"
(for later dynamic estimates).
/* [options] Default value */
{
fullscreen: false,
repeatTesting: 1,
sparkle: true,
pipWidthPx: 208,
showVideo: true,
showFaceOverlay: false,
decimalPlace: 1,
framerate: 3, // Measurement per second
showCancelButton: true,
// Desired distance monitoring
desiredDistanceCm: undefined, // e.g., 60 (which means the target distance is 60 cm)
desiredDistanceTolerance: 1.2, // Tolerable distance would be within [target / 1.2, target * 1.2]
desiredDistanceMonitor: false,
desiredDistanceMonitorCancelable: false,
desiredDistanceMonitorAllowRecalibrate: true,
// Near point
nearPoint: true,
showNearPoint: false,
// Initial measuring
control: true,
headline: "๐ Set up for Distance Tracking",
description: "...",
check: false,
checkCallback: null,
showCancelButton: true,
}
.getDistanceNow([callback])
You can pause active distance tracking, and use this function to get the latest distance at the moment when the user makes reactions. If no callback function is passed in, it will use the one from .trackDistance()
as the default.
The observer's near point is the orthogonal nearest viewing point in the screen, or the plane containing the screen. To track the near point, we assume that the webcam view is orthogonal to the display, and it is placed around 0.5cm above the top center of the screen (e.g., the built-in webcam of a MacBook). Our method is based on the Facemesh model and can give you an approximate estimation of the near point.
Setting nearPoint
option to true
(default) allows the system to track near point and pass the data into the callbackTrack
function along with the distance data. The participant will also be instructed to measure and submit their interpupillary distance before the system can start predict the near point.
The value returned are the horizontal and vertical offsets, in centimeters, compared to the center of the screen. Positive values indicate that the near point is above and to the right of the center point.
- (Beta)
.nudgeDistance(cancelable = false, allowRecalibrate = true, { options, callbackStatic, callbackTrack } = distanceTrackingConfig)
Ifcancelable
istrue
, then participants can use ESC or click to cancel the nudger.distanceTrackingConfig
is used to configure a new distance tracking that can be restarted within the nudger. We recommend to start nudger when starting distance tracking using itsdesiredDistanceCm
anddesiredDistanceMonitor
options. .setDistanceDesired(distanceDesired, [allowedRatio])
.pauseNudger()
.resumeNudger()
.endNudger()
.pauseDistance()
.resumeDistance()
.endDistance([endAll = false])
.showNearPoint([Boolean])
.trackGaze([options, [callbackOnCalibrationEnd, [callbackTrack]]])
Use WebGazer. Pop an interface for participants to calibrate their gaze position on the screen (only when this function is called for the first time), then run in the background and continuously predict the current gaze position. Require access to the camera of the participant's computer. The callback function will be executed repeatedly every time there's a new prediction.
This function should only be called once, unless you want to change the callback functions for every prediction.
Pass { value: { x, y, latencyMs }, timestamp }
(equivalent to RemoteCalibrator.gazePositionPx
) to callbackTrack
function.
/* [options] Default value */
{
fullscreen: false,
// Stop learning and improve the regression model after the calibration process
greedyLearner: false,
// Tracking (predicting) rate per second
framerate: 60,
// Draw the current gaze position on the screen (as a dot)
showGazer: true,
// Show the picture-in-picture video of the participant at the left bottom corner
showVideo: true,
// (Picture in picture) video width in pixels
pipWidthPx: 208,
// Show the face mesh
showFaceOverlay: false,
// How many times participant needs to click on each of the calibration dot
calibrationCount: 1,
// Min accuracy required in degree, set to 'none' to pass the accuracy check
thresholdDeg: 10, ๐ง
decimalPlace: 1, // As the system itself has a high prediction error, it's not necessary to be too precise here
headline: "๐ Set up for Gaze Tracking",
description: "...",
}
.getGazeNow([options, [callback]])
You can pause active gaze tracking after calibration, and use this function to get the latest gaze position at the moment when the user makes reactions, i.e. calling this function in a event listener. This can help reduce computing and get the gaze at the critical moment. If no callback function is passed in, it will use the one from .trackGaze()
as the default.
Pass { value: { x, y }, timestamp }
(equivalent to RemoteCalibrator.gazePositionPx
) to callback. Return the same thing.
/* [options] Default value */
{
wait: 0, // Instead of getting the gaze now, wait for certain milliseconds
frames: 5, // Number of frame samplings used to estimate the current gaze
}
.calibrateGaze([options, [callback]])
Pop an interface for participants to calibrate their gaze position on the screen. Participants need to click on the crosses moving around the screen. This function is automatically called in the .trackGaze()
function when it's called for the first time, but you can always call this function directly as needed, e.g., when the gaze accuracy is low.
/* [options] Default value */
{
greedyLearner: false,
// How many times participant needs to click on each of the calibration dot
calibrationCount: 1,
headline: "๐ Calibrate Gaze",
description: "...",
}
.nudgeGaze([options, [callback]])
Pop an interface for participant to nudge their gaze to the center of the screen. Participants need to click on the crosses at the screen center. Optionally, you can visualize the difference between the lastly estimated gaze and the screen center.
/* [options] Default value */
{
showOffset: true,
}
.pauseGaze()
.resumeGaze()
.endGaze([endAll = false])
.gazeLearning([Boolean, [options]])
WebGazer uses a regression model to always learn and update the model based on the assumption that one would always look at the point where curser makes interaction. However, in a psychophysics experiment, participants may not always look at the place where they click or move the cursor. Thus,greedyLearner
option is set tofalse
by default so that the tracker stops learning after calibration period. But you may also turn it on (or off if needed) again with this function. You can also pass in options, to enable/disable only the click or move event listeners, e.g.,.gazeLearning(true, { click: true, move: false })
will turn on only the listener for the click event. However, it won't turn off for the move events if it's on..showGazer([Boolean])
.showVideo([Boolean])
.showFaceOverlay([Boolean])
Get the setup information of the experiment, including browser type, device model, operating system family and version, etc. See the Getters section for more details.
.backgroundColor()
Set the color of the background. Default#eeeeee
..videoOpacity()
Set the opacity of the video element (intrackDistance
andtrackGaze
). Default0.8
..showCancelButton()
Show the Cancel button or not in the following calibrations. Defaulttrue
.
/* async */ .performance([options, [callback]])
Execute a series of computation demanding tasks for the CPU and GPU, including filling randoms numbers into an array of length of 5000 (repeatedly for a second), generating random numbers (repeatedly for a second), computing for and rendering stressful 2D canvas graphics. Results are reported ascomputeArrayFillMHz
,computeRandomMHz
,idealFps
(canvas FPS without any heavy tasks), andstressFps
in the returned value.options
includes{ testComputation: true, testFrameCount: 180, testObjectCount: 10000, headline, description }
./* async */ .performanceCompute([callback])
GetcomputeArrayFillMHz
,computeRandomMHz
without pulling out any visual elements. Can be called quietly in the background..checkInitialized()
Check if the model is initialized. Return a boolean..getFullscreen()
Get fullscreen mode..newLanguage(lang = 'en-US')
Set a new language for the calibrator.
Get the value directly.
Getters will get null
if no data can be found, i.e. the corresponding function is never called. The values returned by the getter will be wrapped in an object with its corresponding timestamp. Thus, to get the value, add .value
, e.g., RemoteCalibrator.viewingDistanceCm.value
(and use RemoteCalibrator.viewingDistanceCm.timestamp
to get the corresponding timestamp).
.id
The id of the subject. The associated timestamp is the one created at initiation, i.e. wheninit()
is called..displayWidthPx
.displayHeightPx
.windowWidthPx
.windowHeightPx
The display (and window) width and height in pixels..screenWidthCm
.screenHeightCm
.screenDiagonalCm
.screenDiagonalIn
.screenPpi
.screenPhysicalPpi
The physical screen size and monitor PPI in centimeters..viewingDistanceCm
The last measured viewing distance..nearPointCm
The last estimated near point..gazePositionPx
The last measured gaze position on the screen..isFullscreen
Whether the window is in fullscreen mode.
.computeArrayFillMHz
Number of array filling tasks (i.e.Array(5000).fill(Math.floor(Math.random() * 10))
) the computer can finish in a second, in MHz..computeRandomMHz
Number of times the CPU can generate random numbers (i.e.Math.random()
) in a second, in MHz..idealFps
The 2D canvas FPS when there's no tasks at all..stressFps
The 2D canvas FPS when computing for and rendering stressful 2D canvas graphics.
The associated timestamp of the following items is the one created at initiation, i.e. when init()
is called.
.concurrency
The number of cores of CPU. If the browser doesn't support, the value will be-1
.(Removed since 0.6.1.).bot
If the user agent is a bot or not,null
will be returned if no bot detected, e.g.,Googlebot (Search bot) by Google Inc.
..browser
The browser type, e.g.,Safari
,Chrome
..browserVersion
The browser version..deviceType
The type of device, e.g.,desktop
..isMobile
The type of device is mobile or not..model
The model type of the device, e.g.,iPad
..manufacturer
The device manufacturer..engine
The browser engine, e.g.,Webkit
..system
The device operating system, e.g.,OS X 11.2.1 64-bit
..systemFamily
The family name of the device OS, e.g.,OS X
..description
A tidy description of the current environment, e.g.,Chrome 89.0.4389.90 on OS X 11.2.1 64-bit
..fullDescription
The full description of the current environment..userLanguage
The language used in the browser, e.g.,en
.
.language
(e.g.,en-US
,zh-CN
).languageNameEnglish
(e.g.,English
,Chinese (Simplified)
).languageNameNative
(e.g.,็ฎไฝไธญๆ
).languageDirection
(LTR
orRTL
).languagePhraseSource
(e.g.,Denis Pelli & Peiling Jiang 2021.10.10
).supportedLanguages
An array of all supported languages. Can be called before initialization.
Use the following keywords to retrieve the whole dataset.
.displayData
.screenData
.viewingDistanceData
.nearPointData
.PDData
(Inter-pupillary distance data).gazeData
.fullscreenData
.environmentData
.languageData
.version
The RemoteCalibrator version. No associated timestamp.
For building the library locally or development, please follow the steps below.
As we compile Remote Calibrator as a standalone package for distribution, these dependencies are also listed as development dependency in the package.json
configuration.
- animate.css
- leader-line-new
- platform
- react-fast-compare
- sweetalert2
- tinycolor2
- tone
git clone --recurse-submodules https://github.com/EasyEyes/remote-calibrator.git
It is very important to download the submodules with the main repository. If you have cloned the repo without submodules, run the following at the ROOT -
git submodule update --init --recursive
npm run setup
npm start
This will start a local server hosting the example page with hot reload for you to develop. You may access it at localhost:9000
.
npm run build
This command will give you a minimized build of the package output into the lib
folder. You may use the file for production purposes.
- Document changes in CHANGELOG.md.
- Git commit changes.
- Run
npm version v[new.version.number]
(ideally atdevelopment
branch). - Merge into
main
branch, push, and make a release in GitHub.
- Li, Q., Joo, S. J., Yeatman, J. D., & Reinecke, K. (2020). Controlling for participantsโ viewing distance in large-scale, psychophysical online experiments using a virtual chinrest. Scientific reports, 10(1), 1-11.
- Papoutsaki, A., Sangkloy, P., Laskey, J., Daskalova, N., Huang, J., & Hays, J. (2016). Webgazer: Scalable webcam eye tracking using user interactions. In Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence-IJCAI 2016.
As we were wrapping up our development of Remote Calibrator, we realized that Thomas Pronk also made a demonstrative project showing integrating WebGazer into PsychoJS. We would love to acknowledge his work here as well.
- Pronk, T. (2020). Demo of Eye-Tracking via Webcam in PsychoJS (Version 2) [Computer software]. Retrieved from https://gitlab.pavlovia.org/tpronk/demo_eye_tracking2/.