Attributes for Improved Attributes: A Multi-Task Network for Attribute Classification

Attributes, or semantic features, have gained popularity in the past fewyears in domains ranging from activity recognition in video to faceverification. Improving the accuracy of attribute classifiers is an importantfirst step in any application which uses these attributes. In most works todate, attributes have been considered to be independent. However, we know thisnot to be the case. Many attributes are very strongly related, such as heavymakeup and wearing lipstick. We propose to take advantage of attributerelationships in three ways: by using a multi-task deep convolutional neuralnetwork (MCNN) sharing the lowest layers amongst all attributes, sharing thehigher layers for related attributes, and by building an auxiliary network ontop of the MCNN which utilizes the scores from all attributes to improve thefinal classification of each attribute. We demonstrate the effectiveness of ourmethod by producing results on two challenging publicly available datasets.