# Toward a More Ethical Facial Age Estimation: A Generalized Zero-Shot Benchmark Without Training on Children's Data

arXiv: https://arxiv.org/abs/2605.29230 · HTML: https://mlsanity.com/paper/2605.29230/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 17 of its 36 references that are in the map.
- First posted on arXiv 28 May 2026.

## Builds on (most cited first, 17 of 17)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [LAION-5B: An open large-scale dataset for training next generation image-text models](https://mlsanity.com/paper/2210.08402/) (2022)
- [Optuna: A Next-generation Hyperparameter Optimization Framework](https://mlsanity.com/paper/1907.10902/) (2019)
- [Reproducible scaling laws for contrastive language-image learning](https://mlsanity.com/paper/2212.07143/) (2022)
- [Zero-Shot Learning -- A Comprehensive Evaluation of the Good, the Bad and the Ugly](https://mlsanity.com/paper/1707.00600/) (2017)
- [Multimodal datasets: misogyny, pornography, and malignant stereotypes](https://mlsanity.com/paper/2110.01963/) (2021)
- [Feature Generating Networks for Zero-Shot Learning](https://mlsanity.com/paper/1712.00981/) (2017)
- [Large image datasets: A pyrrhic win for computer vision?](https://mlsanity.com/paper/2006.16923/) (2020)
- [Generalized Zero- and Few-Shot Learning via Aligned Variational Autoencoders](https://mlsanity.com/paper/1812.01784/) (2018)
- [Evaluation of Output Embeddings for Fine-Grained Image Classification](https://mlsanity.com/paper/1409.8403/) (2014)
- [Rank consistent ordinal regression for neural networks with application to age estimation](https://mlsanity.com/paper/1901.07884/) (2019)
- [Deep Label Distribution Learning with Label Ambiguity](https://mlsanity.com/paper/1611.01731/) (2016)
- [Deep Neural Networks for Rank-Consistent Ordinal Regression Based On Conditional Probabilities](https://mlsanity.com/paper/2111.08851/) (2021)
- [A Call to Reflect on Evaluation Practices for Age Estimation: Comparative Analysis of the State-of-the-Art and a Unified Benchmark](https://mlsanity.com/paper/2307.04570/) (2023)
- [#PraCegoVer: A Large Dataset for Image Captioning in Portuguese](https://mlsanity.com/paper/2103.11474/) (2021)
- [Neglected Risks: The Disturbing Reality of Children's Images in Datasets and the Urgent Call for Accountability](https://mlsanity.com/paper/2504.14446/) (2025)
- [Minimizing Risk Through Minimizing Model-Data Interaction: A Protocol For Relying on Proxy Tasks When Designing Child Sexual Abuse Imagery Detection Models](https://mlsanity.com/paper/2505.06621/) (2025)

## Abstract (arXiv)

Age estimation from facial images typically relies on training data that includes images of minors, a practice that raises ethical, legal, and privacy concerns and that child-data governance frameworks explicitly advise against. While the task remains relevant (e.g., for detecting child sexual abuse imagery), we advocate against using data from minors entirely and quantify what the exclusion costs in accuracy. We formalize age estimation without children's training data as a generalized zero-shot learning (GZSL) problem: age intervals present during training are seen classes and withheld intervals are unseen, with models evaluated jointly on both. The generalized setting, rather than conventional zero-shot evaluation on unseen classes alone, is the appropriate one here because a deployed estimator must operate across the entire lifespan, not only on the interval withheld from it. Revisiting six widely used datasets, we introduce standardized splits with strict age-group separation. For datasets with identity annotations, subject-age-exclusive splits prevent identity leakage across the seen/unseen boundary. Evaluating nine state-of-the-art age estimation methods under this protocol reveals that all of them fail to generalize to unseen age groups, suffering substantial degradation --- on average 46.4%, and up to 52.8% --- relative to the supervised baseline. Moreover, models do not simply degrade: they systematically anchor predictions for unseen ages to nearby seen classes, a manifestation of the well-known seen-class bias in generalized zero-shot learning.