Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Eye Feel You: A DenseNet-driven User State Prediction Approach

Kamrul Hasan, Oleg V. Komogortsev

TL;DR

The paper tackles predicting subjective states from objective eye-tracking by formulating it as a multi-target regression problem and proposing a pre-activation DenseNet-based regressor that operates on gaze velocity signals. The approach reduces reliance on handcrafted features and demonstrates robust cross-round (within-subject) performance improvements, though cross-subject generalization remains modest, highlighting the need for personalization or domain adaptation. Two complementary experiments assess longitudinal stability and robustness to new individuals using the GazeBase dataset, revealing meaningful within-subject gains but limited out-of-subject transfer. The work advances understanding of how oculomotor dynamics reflect subjective experience over time and suggests practical directions for personalized gaze-based state estimation in real-world deployments.

Abstract

Subjective self-reports, collected with eye-tracking data, reveal perceived states like fatigue, effort, and task difficulty. However, these reports are costly to collect and challenging to interpret consistently in longitudinal studies. In this work, we focus on determining whether objective gaze dynamics can reliably predict subjective reports across repeated recording rounds in the eye-tracking dataset. We formulate subjective-report prediction as a supervised regression problem and propose a DenseNet-based deep learning regressor that learns predictive representations from gaze velocity signals. We conduct two complementary experiments to clarify our aims. First, the cross-round generalization experiment tests whether models trained on earlier rounds transfer to later rounds, evaluating the models' ability to capture longitudinal changes. Second, cross-subject generalization tests models' robustness by predicting subjective outcomes for new individuals. These experiments aim to reduce reliance on hand-crafted feature designs and clarify which states of subjective experience systematically appear in oculomotor behavior over time.

Eye Feel You: A DenseNet-driven User State Prediction Approach

TL;DR

The paper tackles predicting subjective states from objective eye-tracking by formulating it as a multi-target regression problem and proposing a pre-activation DenseNet-based regressor that operates on gaze velocity signals. The approach reduces reliance on handcrafted features and demonstrates robust cross-round (within-subject) performance improvements, though cross-subject generalization remains modest, highlighting the need for personalization or domain adaptation. Two complementary experiments assess longitudinal stability and robustness to new individuals using the GazeBase dataset, revealing meaningful within-subject gains but limited out-of-subject transfer. The work advances understanding of how oculomotor dynamics reflect subjective experience over time and suggests practical directions for personalized gaze-based state estimation in real-world deployments.

Abstract

Subjective self-reports, collected with eye-tracking data, reveal perceived states like fatigue, effort, and task difficulty. However, these reports are costly to collect and challenging to interpret consistently in longitudinal studies. In this work, we focus on determining whether objective gaze dynamics can reliably predict subjective reports across repeated recording rounds in the eye-tracking dataset. We formulate subjective-report prediction as a supervised regression problem and propose a DenseNet-based deep learning regressor that learns predictive representations from gaze velocity signals. We conduct two complementary experiments to clarify our aims. First, the cross-round generalization experiment tests whether models trained on earlier rounds transfer to later rounds, evaluating the models' ability to capture longitudinal changes. Second, cross-subject generalization tests models' robustness by predicting subjective outcomes for new individuals. These experiments aim to reduce reliance on hand-crafted feature designs and clarify which states of subjective experience systematically appear in oculomotor behavior over time.
Paper Structure (18 sections, 2 equations, 1 figure, 4 tables)

This paper contains 18 sections, 2 equations, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Eye Tracking as an Objective Signal for Interactive Systems
  4. Subjective State Measurement and Gaze-Based Prediction
  5. Learning-Based Regression for Subjective-State Prediction
  6. Proposed Architecture
  7. Data Preprocessing
  8. DenseNet Feature Extractor and Regression Head
  9. Experiment
  10. Hardware & Software
  11. Dataset
  12. Training
  13. Testing
  14. Results
  15. Performance on Cross-rounds
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Overview of the architecture, where the positional signal $p$ is processed by the Preprocessor, yielding a velocity signal, which is then fed to the Pre-activation DenseNet. Here, each convolutional layer has a kernel size $k = 3$, a stride $s = 1$, and a varying dilation rate $d$. Later, Regression Head takes the output from DenseNet and predicts Subject Scores, where $N$ denotes the number of predicted scores.