Research on Face Attribute Recognition Based on Multi-task CNN Network

In recent years, the research on face attribute recognition has received special attention due to its research and practical value. Face attribute recognition with multi-task convolutional neural network achieves higher reliability and accuracy than traditional feature extraction methods. In this paper, we propose a hard parameter sharing-channel split network(HPS-CS) to enhance channel information sharing and reduce the number of parameters for the face attribute recognition task. HPS-CS is employed in this work to identify 7 types of face attributes on the CelebA dataset. The average accuracy for the experiment is 94.59% and the model size is only 2.88 MiB. The competitive result proves that the network has advantages in multi-task face attribute recognition.