Saxena-AVS-2026

Studying visual attention beyond the lab: Lessons from applied eye-tracking

Shreshth Saxena (saxens17@mcmaster.ca) & Dr. Lauren Fink

Studying visual attention beyond the lab: Lessons from applied eye-tracking

Shreshth Saxena (saxens17@mcmaster.ca) & Dr. Lauren Fink

Tracking attention beyond the lab with eye movements

How do we run eye-tracking studies outside the lab?

Remote settings
eyelink

In-person settings
tasks

Remote settings

Relying on state-of-the-art hardwaresoftware

Study 1: Reliably calibrating screen-camera setups
Study 2: Characterising the accuracy of webcam eye-tracking
Study 3: Optimising the speed-accuracy trade-off

Study 1: Reliably calibrating screen-camera setups

Saxena et al. 2022. Towards efficient calibration for webcam eye-tracking in online experiments. Symposium on Eye Tracking Research and Applications (ETRA '22) . Association for Computing Machinery. https://doi.org/10.1145/3517031.3529645

Study 2: Characterising the accuracy of webcam eye-tracking

Saxena et al. 2023. Deep learning models for webcam eye tracking in online experiments. Behaviour Research Methods. https://doi.org/10.3758/s13428-023-02190-6

Study 3: Optimising the speed-accuracy trade-off

Deep learning outperforms present SOTA, but offline inference limits real-time use and privacy.

Study 3: Optimising the speed-accuracy trade-off

Saxena and Fink. 2026. WebSight: Accessible and privacy-aware webcam eye tracking using real-time deep learning in the browser. Manuscript in Preparation

Lessons from eye-tracking in remote settings

Calibration sequence (tasks and placement) strongly affects final prediction accuracy
Traditional event-detection and analysis pipelines do not work for 30Hz eye-tracking data
Real-time deep learning inference in the web-browser is still in beta

Web development shifted back to server-side compute with the rise of LLMs
WebGPU and WebONNX suffer from abondoned research and development
WebSight works around it using classic web workers and webGL acceleration!

Proprietary, GUI-based experiment builders don’t scale

Problem no. 1: Openwashing (https://en.wikipedia.org/wiki/Openwashing)
Problem no. 2: Working with a limited toolset and inconsistent terminologies

Saxena et al. 2026. AVOKE: an open-source toolbox for audiovisual web experiments in jsPsych. PeerJ.
Code and demo available at https://beatlab-mcmaster.github.io/AVOKE

How do we take eye tracking outside the lab?

Remote settings
eyelink

In-person settings
tasks

In-person settings

Relying on hardware and software

x N

Relying on hardware and software

Proprietary firmware/software
Logistical challenges when managing a fleet
Synchronising data from multiple sensors and clocks
Synchronising data from multiple egocentric perspectives
Joint attention visualisation and analysis

Study 4: Scaling Mobile Eye-tracking

Proprietary firmware/software
Logistical challenges when managing a fleet
Synchronising data from multiple sensors and clocks
Synchronising data from multiple egocentric perspectives
Joint attention visualisation and analysis

Saxena et al. 2025. SocialEyes: Scaling Mobile Eye-tracking to Multi-person Social Settings. ACM SIGCHI 2025. https://doi.org/10.1145/3706598.3713910

Study 4: Scaling Mobile Eye-tracking

Proprietary firmware/software
Logistical challenges when managing a fleet
Synchronising data from multiple sensors and clocks
Synchronising data from multiple egocentric perspectives
Joint attention visualisation and analysis

Saxena et al. 2025. SocialEyes: Scaling Mobile Eye-tracking to Multi-person Social Settings. ACM SIGCHI 2025. https://doi.org/10.1145/3706598.3713910

Study 4: Scaling Mobile Eye-tracking

Proprietary firmware/software
Logistical challenges when managing a fleet
Synchronising data from multiple sensors and clocks
Synchronising data from multiple egocentric perspectives
Joint attention visualisation and analysis

Saxena et al. 2025. SocialEyes: Scaling Mobile Eye-tracking to Multi-person Social Settings. ACM SIGCHI 2025. https://doi.org/10.1145/3706598.3713910

Study 4: Scaling Mobile Eye-tracking

Proprietary firmware/software
Logistical challenges when managing a fleet
Synchronising data from multiple sensors and clocks
Synchronising data from multiple egocentric perspectives
Joint attention visualisation and analysis

Saxena et al. 2025. SocialEyes: Scaling Mobile Eye-tracking to Multi-person Social Settings. ACM SIGCHI 2025. https://doi.org/10.1145/3706598.3713910

Study 4: Scaling Mobile Eye-tracking

Proprietary firmware/software
Logistical challenges when managing a fleet
Synchronising data from multiple sensors and clocks
Synchronising data from multiple egocentric perspectives
Joint attention visualisation and analysis

Saxena et al. 2025. SocialEyes: Scaling Mobile Eye-tracking to Multi-person Social Settings. ACM SIGCHI 2025. https://doi.org/10.1145/3706598.3713910

Study 4: Scaling Mobile Eye-tracking

Proprietary firmware/software
Logistical challenges when managing a fleet
Synchronising data from multiple sensors and clocks
Synchronising data from multiple egocentric perspectives
Joint attention visualisation and analysis

Saxena et al. 2025. SocialEyes: Scaling Mobile Eye-tracking to Multi-person Social Settings. ACM SIGCHI 2025. https://doi.org/10.1145/3706598.3713910

Study 4: Scaling Mobile Eye-tracking

Saxena et al. 2025. SocialEyes: Scaling Mobile Eye-tracking to Multi-person Social Settings. ACM SIGCHI 2025. https://doi.org/10.1145/3706598.3713910

Lessons from eye-tracking in co-located, social settings

Inherent trade-off between non-invasive design vs. experimental control
Manual annotations and synchronisation don't scale
Mobile eye-trackers are mostly built/tested for single-user, self-operation
The cost of maintaining a fleet of eye-tracking hardware quickly adds up
Proprietary softwares and subscriptions make it much worse
Call to researchers, developers, and manufacturers:

Saxena et al. 2026. OpenET: Converging on open standards for hardware-interfacing and data-sharing in eye-tracking.
28th International Conference on Human-Computer Interaction, Montreal, Canada.
https://openet-initiative.github.io/

Lessons from conducting eye-tracking beyond the lab

Standardization and open-source development is the key

community-contributions have solved similar challenges in other spaces

Wide range of applications: cognitive neuroscience, healthcare, broadcasting, education, etc.

I benefited greatly from interdisciplinary collaboration

Method development requires multiple iterations and continued maintenance

Can take a full PhD and more

Thank you for your attention

Acknowledgements to all contributors, collaborators, and volunteers from

Why take eye tracking beyond the lab?

Study social attention in context
Study cross-modal dynamics with complex, ecological stimuli
Reach global and underrepresented populations
Bridge research and deployment–align methods with real-world systems and industry use cases
Findings from lab based studies often do not replicate in naturalistic/real-world experimentation

Foulsham et al. (2011). The where, what and when of gaze allocation in the lab and the natural environment. Vision Research, doi: 10.1016/j.visres.2011.07.002.
Wohltjen & Wheatley (2024) Interpersonal eye-tracking reveals the dynamics of interacting minds. Front. Hum. Neurosci. doi: 10.3389/fnhum.2024.1356680
Laborde et al. (2025) Vision toolkit part 1. Neurophysiological foundations and experimental paradigms in eye-tracking research: a review. Front Physiol. doi: 10.3389/fphys.2025.1571534.
Hoppe et al. (2018) Eye Movements During Everyday Behavior Predict Personality Traits. Front. Hum. Neurosci. doi: 10.3389/fnhum.2018.00105

Summary of results

Distributed (bookended) and hybrid calibration outperforms sequential calibration
Gaze prediction accuracy differs across eye-movement and task types
Deep learning reduces calibration and improves reliability over longer trials
Fixation accuracy of ~2.4° from best performing model; surpasses previous state-of-the-art
Prediction accuracy drops at edges and bottom regions of the screen
Appearance-based methods match high-resolution trackers for blink detection
Currently running experiments with real-time deep learning model in web-browsers; to be released as a jsPsych extension

Study 4: Scaling Mobile Eye-tracking

Proprietary firmware/software
Logistical challenges when managing a fleet
Synchronising data from multiple sensors and clocks
Synchronising data from multiple egocentric perspectives
Joint attention visualisation and analysis

Saxena et al. 2025. SocialEyes: Scaling Mobile Eye-tracking to Multi-person Social Settings. ACM SIGCHI 2025. https://doi.org/10.1145/3706598.3713910

AVOKE demo

OpenET demo