Eye-tracking and AI: Unveiling the Limitations and Synergies
summary
Company-built webcam-based remote eye tracking and a portable lab tracker (Essa/Tobii) are compared against two popular predictive saliency models across packaging, webpages, virtual shelves and 18 video clips. Predictive models often highlight high-contrast color or faces but systematically underweight semantically loaded regions like brand names, miss reading order and time-frame dynamics, and disagree with each other. A combined workflow—saliency prediction corrected with real eye-data and industry-specific model tuning—is scalable and improves accuracy; transparency about model bias is essential.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:46Market research and AI evolution
- 3:41AI in eye-tracking tech
- 6:07Heat maps vs AI predictions
- 10:47Testing AI and eye-tracking accuracy
- 13:38AI limitations in marketing
- 15:44Future of AI and eye-tracking
- 16:47Q&A
related talks
20mBuilding a Personalized, Auto-Calibrating Eye Tracker from User Interactions
48mEye Tracking (a Deep Learning Approach)
21mSelf-Calibrating Head-Mounted Eye Trackers Using Egocentric Visual Saliency
19mEveryday Eye Contact Detection Using Unsupervised Gaze Target Discovery
17mEyeGrip: Detecting Targets in a Series of Uni-directional Moving Objects Using Optokinetic ...
19mEyeScout: Active Eye Tracking for Position and Movement Independent Gaze Interaction with Large ...
19mGazeProjector: Accurate Gaze Estimation and Seamless Gaze Interaction Across ...
06mMemX: An Attention-Aware Smart Eyewear System for Personalized Moment Auto-capture