[2603.24724] Is Geometry Enough? An Evaluation of Landmark-Based Gaze Estimation

Computer Science > Computer Vision and Pattern Recognition arXiv:2603.24724 (cs) [Submitted on 25 Mar 2026] Title:Is Geometry Enough? An Evaluation of Landmark-Based Gaze Estimation Authors:Daniele Agostinelli, Thomas Agostinelli, Andrea Generosi, Maura Mengoni View a PDF of the paper titled Is Geometry Enough? An Evaluation of Landmark-Based Gaze Estimation, by Daniele Agostinelli and 3 other authors View PDF HTML (experimental) Abstract:Appearance-based gaze estimation frequently relies on deep Convolutional Neural Networks (CNNs). These models are accurate, but computationally expensive and act as "black boxes", offering little interpretability. Geometric methods based on facial landmarks are a lightweight alternative, but their performance limits and generalization capabilities remain underexplored in modern benchmarks. In this study, we conduct a comprehensive evaluation of landmark-based gaze estimation. We introduce a standardized pipeline to extract and normalize landmarks from three large-scale datasets (Gaze360, ETH-XGaze, and GazeGene) and train lightweight regression models, specifically Extreme Gradient Boosted trees and two neural architectures: a holistic Multi-Layer Perceptron (MLP) and a siamese MLP designed to capture binocular geometry. We find that landmark-based models exhibit lower performance in within-domain evaluation, likely due to noise introduced into the datasets by the landmark detector. Nevertheless, in cross-domain evaluation, the proposed M...