GroupFace: Imbalanced Age Estimation Based on Multi-hop Attention Graph Convolutional Network and Group-aware Margin Optimization

TL;DR

GroupFace addresses the persistent challenge of imbalanced age estimation by jointly learning discriminative group-aware facial representations and optimizing margins for long-tailed age groups. It introduces EMAGCN, which blends multi-hop attention diffusion with adaptive decay, message dropping, residuals, and power-iteration acceleration to capture local and global aging cues. A reinforcement learning–driven dynamic group-aware margin optimization module learns group-specific margins via a Markov decision process, balancing inter-class separability and intra-class proximity. Across MORPH-II, UTK-Face, CACD, ChaLearn LAP 2015, and MIVIA, GroupFace achieves strong overall accuracy while improving balanced generalization for tail groups, outperforming or matching state-of-the-art methods with markedly fewer parameters. These results demonstrate practical impact for fairer age estimation in real-world, long-tailed datasets and suggest fruitful directions with language-image pre-training.

Abstract

With the recent advances in computer vision, age estimation has significantly improved in overall accuracy. However, owing to the most common methods do not take into account the class imbalance problem in age estimation datasets, they suffer from a large bias in recognizing long-tailed groups. To achieve high-quality imbalanced learning in long-tailed groups, the dominant solution lies in that the feature extractor learns the discriminative features of different groups and the classifier is able to provide appropriate and unbiased margins for different groups by the discriminative features. Therefore, in this novel, we propose an innovative collaborative learning framework (GroupFace) that integrates a multi-hop attention graph convolutional network and a dynamic group-aware margin strategy based on reinforcement learning. Specifically, to extract the discriminative features of different groups, we design an enhanced multi-hop attention graph convolutional network. This network is capable of capturing the interactions of neighboring nodes at different distances, fusing local and global information to model facial deep aging, and exploring diverse representations of different groups. In addition, to further address the class imbalance problem, we design a dynamic group-aware margin strategy based on reinforcement learning to provide appropriate and unbiased margins for different groups. The strategy divides the sample into four age groups and considers identifying the optimum margins for various age groups by employing a Markov decision process. Under the guidance of the agent, the feature representation bias and the classification margin deviation between different groups can be reduced simultaneously, balancing inter-class separability and intra-class proximity. After joint optimization, our architecture achieves excellent performance on several age estimation benchmark datasets.

Figures (10)

• Figure 1: The illustration of age estimation with class imbalanced learning. Most face datasets have an imbalanced distribution of race and age groups, such that the recognition bias is high for the long-tailed groups. Our GroupFace can achieve a balanced generalization capability for different age groups by discriminative feature extraction and group-aware margin optimization.
• Figure 2: The overall framework of our proposed imbalanced learning method GroupFace. Face images are segmented into patches as nodes, and then a multi-hop attention graph convolutional network will fuse global and local information to model deep facial aging for capturing the discriminative features of different groups. Through joint optimization, the dynamic group-aware margin strategy based on reinforcement learning will identify the optimum margins for various age groups to mitigate the bias of imbalanced learning.
• Figure 3: The illustration of the main designs of EMAGCN to capture discriminative features fusing global and local information.
• Figure 4: The illustration of EMAGCN blocks. In the schematic above, the relations between multi-hops are obtained by attention diffusion.
• Figure 5: The illustration of different margin losses. GroupFace employs the dynamic group-aware margin loss, which can adaptively provide suitable and unbiased margins for different age groups.