| 摘 要: 在人脸年龄估计任务中,针对现有深度学习模型在提取层次化特征方面存在不足的问题,提出了一种基于Swin Transformer的层次化年龄估计网络(HAEN-Swin)。该网络包含浅层和深层特征提取模块,分别捕获面部基础结构和细节特征,实现多层次特征提取,显著提升年龄估计准确性。此外,针对数据集样本分布不均衡的问题,网络还引入了Dice相似系数作为损失函数的一部分。实验结果表明,HAEN-Swin在MORPH数据集上表现最佳,平均绝对误差(MAE)为2.45,充分验证了该模型的有效性和优越性。 |
| 关键词: 年龄估计 SwinTransformer Dice相似系数 |
|
中图分类号: TP391
文献标识码: A
|
|
| Hierarchical Age Estimation Network Based on Swin Transformer |
|
XU Linbu, HU Chunlong
|
(School of Computer Science, Jiangsu University of Science and Technology, Zhenjiang 212114, China)
xulinbu1012@163.com; huchunlong@just.edu.cn
|
| Abstract: In facial age estimation tasks, addressing the limitations of existing deep learning models in extracting hierarchical features, this paper proposes a Hierarchical Age Estimation Network based on Swin Transformer (HAEN-Swin). The network incorporates shallow and deep feature extraction modules to capture fundamental facial structures and detailed characteristics respectively, achieving multi-level feature extraction and significantly enhancing age estimation accuracy. Furthermore, to mitigate dataset imbalance issues, the network integrates the Dice similarity coefficient as part of the loss function. Experimental results demonstrate that HAEN-Swin achieves optimal performance on the MORPH dataset with a Mean Absolute Error (MAE) of 2.45, fully validating the model’s effectiveness and superiority. |
| Keywords: age estimation Swin Transformer Dice similarity coefficient |
