Michelle Green, Ph.D.

Assistant Professor of Psychology
Barnard College of Columbia University
 

The Visual Experience Dataset: Insights into Active Visual Sampling and the Statistics of our Visual Diet

In this talk, I will introduce the Visual Experience Dataset, a 244-hour mobile eyetracking dataset that captures real-world visual experiences across diverse environments and activities. By combining head and eye movements with high-resolution video, this dataset provides a unique opportunity to explore how humans actively sample visual information in their everyday surroundings. I will share some new insights from the dataset, including analyzing gaze-centered image statistics, gaze prediction, and the strength of scene-object context. These findings show how we direct our gaze, navigate complex environments, and process visual information in real-world contexts. Additionally, I will discuss the implications of this work for training and using computer vision models, highlighting the challenges and biases that arise when comparing real-world visual experiences with internet-derived image datasets.