Data
Abstract
Data set bias not only compromises the fairness, accuracy and effectiveness of trained models, but also leads to a lower performance in real-world scenarios compared to the evaluation results obtained with a specific data set. This issue is especially evident in the estimation of head pose, as current data sets suffer from a limited number of images, imbalanced data distributions, the high cost of annotation, and ethical concerns. Synthetic data offers a promising solution to address these challenges, but current semi-synthetic data sets fail to deliver satisfactory results, likely due to the limited realism of the generated faces and the heavily skewed pose distribution. In this paper, we report the existence of data set biases in the most widely used head pose estimation benchmarks, which lead to an optimistic estimation of model performance in real-world scenarios. To mitigate this issue, we create a synthetic image data set using a generative model with explicit control over the head pose. Our experiments demonstrate that incorporating our synthetic images leads to improved generalization and accuracy.
Figure 3: Representative face samples of HPGEN
Citation
Roberto Valle and José Miguel Buenaposada and Luis Baumela. Reducing Head Pose Estimation Data Set Bias With Synthetic Data. IEEE Access 13: 73530-73539 (2025)
@article{Valle25,
author = {Roberto Valle and Jos{\'{e}} Miguel Buenaposada and Luis Baumela},
title = {Reducing Head Pose Estimation Data Set Bias With Synthetic Data},
journal = {{IEEE} Access},
volume = {13},
pages = {73530-73539},
year = {2025},
url = {https://doi.org/10.1109/ACCESS.2025.3561506}
}