Gender and Age Estimation using Transfer Learning with Multi-tasking Approach

The estimation of Age and Gender of a Human being using facial characteristic is getting lot of attractions these days due to wide variety of its applications in the real-world scenarios. Significant work has been carried out in this regard and various methods and approaches has been published which have offered very good results as far as accuracy are concerned. This paper brings out a comparison of most famous pretrained models for estimating both gender and age of a person in videos using multi-tasking approach enabling a single model to estimate both age as well as gender. The estimation accuracy of EfficientNetV2B1 model using multi-tasking approach provides good accuracy (of 90.31) and MAE (of 0.063) compared to other models. The publicly available dataset UTKFACE has been used for training the multi-tasking CNN.