Synthetic Infrared Eye Images for Scalable Gaze Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Collecting real-world data for training machine learning models, particularly for gaze estimation under infrared light conditions, is tedious and inefficient.
Innovation Solution
A simulation engine generates synthetic infrared images of human eyes with accurate anatomical features and rendering settings to simulate real-world conditions, accompanied by labels for training neural networks to estimate gaze accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If real-world data is collected for training machine learning models, then the model can learn from actual conditions, but the data collection process is tedious and inefficient
Solution Approach 1:
The patent generates synthetic infrared images that copy and simulate real-world eye appearance and infrared reflection characteristics. These synthetic images replicate the optical properties and anatomical features of real eyes under infrared illumination, providing training data that mirrors actual conditions without requiring physical data collection.
Solution Approach 2:
The system varies multiple parameters including eye anatomy, lighting conditions, camera settings, and infrared reflection properties to generate diverse synthetic training data. This parameter variation enables the model to learn across different scenarios efficiently, replacing the need to collect data from numerous real-world conditions.
2Measurement precision
If real-world infrared images are collected for gaze estimation, then accurate training data is obtained, but the process is time-consuming and difficult to scale
Solution Approach 1:
The synthetic images are pre-generated with known ground truth annotations for eye features and gaze directions before model training begins. This preliminary preparation of training data eliminates the time-consuming process of collecting and annotating real-world images, allowing immediate use in model training.
Solution Approach 2:
The system automatically generates synthetic training data without requiring manual intervention for data collection or annotation. The synthetic image generation process self-provides both the training images and their corresponding labels, eliminating human time investment in data preparation while maintaining training accuracy.
Data Source
AI summary
One embodiment of a method includes calculating one or more activation values of one or more neural networks trained to infer eye gaze information based, at least in part, on eye position of one or more images of one or more faces indicated by an infrared light reflection from the one or more images.