AB-net: Adult- Baby Net

The problem of estimating and classifying age from a given input im- age is age old. With the advancement of modern technology and recent progress in the field of deep learning it has been made possible to gain success for this particular application. We present a novel algorithm inspired by the Multi-Task Cascaded Convolutional Neural Networks architecture and a subsequent pipeline to a Prediction architecture thus defining an end – to – end pipeline, which can predict with great certainty the age group, from a given input image. We introduce approaches in data preparation techniques such as Complex Negatives and IOU based segregation which proves beneficial to the reduction of false-positive classification of samples. Thus significantly proving on benchmarks to achieve robust performance with regard to the age classification problem. The proposed method is able to accurately predict and has been benchmarked on the FDDB dataset [7].

Keywords

• Face detection

• Face classification

• Adult-Baby detection