FOSS: Multi-Person Age Estimation with Focusing on Objects and Still Seeing Surroundings
Age estimation from images can be used in many practical scenes. Most of the previous works targeted on the estimation from images in which only one face exists. Also, most of the open datasets for age estimation contain images like that. However, in some situations, age estimation in the wild and for multi-person is needed. Usually, such situations were solved by two separate models; one is a face detector model which crops facial regions and the other is an age estimation model which estimates from cropped images. In this work, we propose a method that can detect and estimate the age of multi-person with a single model which estimates age with focusing on faces and still seeing surroundings. Also, we propose a training method which enables the model to estimate multi-person well despite trained with images in which only one face is photographed. In the experiments, we evaluated our proposed method compared with the traditional approach using two separate models. As the result, the accuracy could be enhanced with our proposed method. We also adapted our proposed model to commonly used single person photographed age estimation datasets and it is proved that our method is also effective to those images and outperforms the state of the art accuracy.