Abstract
Learning generic and robust feature representations withdata from multiple domains for the same problem is ofgreat value, especially for the problems that have multiple datasets but none of them are large enough to pro-vide abundant data variations. In this work, we present apipeline for learning deep feature representations from mul-tiple domains with Convolutional Neural Networks (CNNs).When training a CNN with data from all the domains, someneurons learn representations shared across several domains, while some others are effective only for a specificone. Based on this important observation, we propose a Domain Guided Dropout algorithm to improve the feature learning procedure. Experiments show the effectiveness of our pipeline and the proposed algorithm. Our methodson the person re-identification problem outperform state-of-the-art methods on multiple datasets by large margins.