Reproduce “PupilNet: Convolutional Neural Networks for Robust Pupil Detection” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Wolfgang Fuhl, Thiago Santini, Gjergji Kasneci, Enkelejda Kasneci

PupilNet introduces a dual convolutional neural network pipeline that significantly enhances pupil detection in challenging, natural environments. It achieves approximately 25% higher detection rates compared to prior state-of-the-art methods, while maintaining real-time performance on standard mobile CPU systems.

Key insights

  • A dual-stage CNN architecture effectively combines robustness in coarse localization with precision in fine-grained detection, outperforming single-stage and heuristic-based methods in complex scenarios.
  • CNNs can autonomously learn fundamental image processing features, such as center-surround differences and edge detectors, crucial for robust pupil identification in diverse conditions.
  • The design enables real-time operation on conventional mobile computing platforms by strategically allocating computational resources across two stages, fulfilling the requirement for pervasive eye-tracking without dedicated GPUs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.