Stage-wise Adaptive Label Distribution for Facial Age Estimation

2025-08-30

AI Summary

In facial age estimation, age labels exhibit significant and stage-dependent ambiguity, which existing methods often overlook due to their failure to model structural differences across age stages. This paper is the first to identify and formalize the stage-wise regularity of age ambiguity, proposing a stage-adaptive label distribution learning framework. Specifically, we analyze embedding-space similarity to uncover age-specific ambiguity patterns, introduce stage-adaptive variance modeling to characterize uncertainty, and design a weighted loss function for optimized distribution learning. Evaluated on MORPH-II and FG-NET, our method achieves state-of-the-art MAEs of 1.74 and 2.15, respectively. The core contribution lies in revealing and exploiting the intrinsic stage-wise nature of age ambiguity—enabling more robust, interpretable, and principled age estimation.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Intelligent Robots: State Estimation
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Label ambiguity poses a significant challenge in age estimation tasks. Most existing methods address this issue by modeling correlations between adjacent age groups through label distribution learning. However, they often overlook the varying degrees of ambiguity present across different age stages. In this paper, we propose a Stage-wise Adaptive Label Distribution Learning (SA-LDL) algorithm, which leverages the observation -- revealed through our analysis of embedding similarities between an anchor and all other ages -- that label ambiguity exhibits clear stage-wise patterns. By jointly employing stage-wise adaptive variance modeling and weighted loss function, SA-LDL effectively captures the complex and structured nature of label ambiguity, leading to more accurate and robust age estimation. Extensive experiments demonstrate that SA-LDL achieves competitive performance, with MAE of 1.74 and 2.15 on the MORPH-II and FG-NET datasets.

Problem

• Addresses label ambiguity in facial age estimation
• Models stage-wise patterns of label ambiguity
• Improves accuracy through adaptive variance and weighted loss

Innovation

• Stage-wise adaptive label distribution learning
• Adaptive variance modeling and weighted loss
• Embedding similarity analysis for ambiguity patterns