Enhancing Age and Gender Classification through cGAN-based Data Augmentation

Age and gender classification from facial images have wide-ranging applications. Deep learning has revolutionized these tasks, outperforming traditional machine learning methods. However, existing studies have focused on individual models, leading to a performance gap and a need for alternative approaches. In this paper, we propose a data augmentation method for age and gender classification using geometric augmentation, an age transformation model, and a filtering system. Our approach utilizes a conditional Generative Adversarial Network (cGAN) with cropping and flipping techniques to augment the dataset, enhancing its diversity. We also introduce a filtering system to improve the reliability of the age and gender classification model by ensuring the accuracy of labeled synthesized data. We evaluate our method using customized VGG16 and ConvNeXt models and observe significant improvements in age and gender classification tasks.