Gaze estimation problem tackled through synthetic images

TL;DR

This paper evaluates a synthetic image framework for gaze estimation, demonstrating its robustness and potential for calibration, with comparable or better performance than real data-based methods.

Contribution

It introduces a synthetic environment for gaze estimation that achieves stable results and shows promise for user-specific calibration strategies.

Findings

Synthetic framework yields robust gaze estimation performance.

Models pretrained on synthetic data perform well in calibration.

Synthetic images provide comparable or better results than real data.

Abstract

In this paper, we evaluate a synthetic framework to be used in the field of gaze estimation employing deep learning techniques. The lack of sufficient annotated data could be overcome by the utilization of a synthetic evaluation framework as far as it resembles the behavior of a real scenario. In this work, we use U2Eyes synthetic environment employing I2Head datataset as real benchmark for comparison based on alternative training and testing strategies. The results obtained show comparable average behavior between both frameworks although significantly more robust and stable performance is retrieved by the synthetic images. Additionally, the potential of synthetically pretrained models in order to be applied in user's specific calibration strategies is shown with outstanding performances.