Eye Tracking (a Deep Learning Approach)
summary
A practical walkthrough of building an eye-tracking system using deep learning: collecting and labeling training data, engineering the preprocessing and model iterations, and integrating the model into a usable product that lets users control the mouse with their gaze. The talk emphasizes the many small engineering subtasks and pitfalls that affect accuracy and shows the gap between theoretical models and real-world usability, arguing that pragmatic engineering trade-offs can yield satisfactory performance even if not perfect.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 2:27Eye tracking and uses
- 11:24Game data generation
- 15:16Image processing and facial points
- 20:17Deep learning model for eye tracking
- 23:17Testing and practical apps
- 27:48Lessons learned and biases
- 32:03User experience and expectations
- 35:56Conclusions and recommendations
- 40:09Q&A
related talks
20mBuilding a Personalized, Auto-Calibrating Eye Tracker from User Interactions
40mThe eye of the Python, an eye tracking system. From zero to...
20mEye-tracking and AI: Unveiling the Limitations and Synergies
19mEyeScout: Active Eye Tracking for Position and Movement Independent Gaze Interaction with Large ...
19mEveryday Eye Contact Detection Using Unsupervised Gaze Target Discovery
38mWhere are we looking? Predicting human gaze using deep networks.
21mSelf-Calibrating Head-Mounted Eye Trackers Using Egocentric Visual Saliency
10mHAGI: Head-Assisted Gaze Imputation for Mobile Eye Trackers