Fine-tunning and knowledge transfer for gender recognition using CNN
T. Varga · Tesis Doctorals en Xarxa (Consorci de Serveis Universitaris de Catalunya) · 2016
Automatic gender classification has gain popularity particularly with the increasing amount of social platforms and social media providers. Also, recently has been noted an increase use of CNNs to solve different classification tasks, one of which is gender recognition. Training and testing a CNN often involves a large amount of time and the classification results are not consistent across databases. In this work I use a pre-trained CNN with one data set, retrain it in order to obtain similar result when classifying another data set with much less effort involve in training. I evaluate the method using the Image of Group data set with a CNN trained with Adience data set.