Cross-dataset Visual Recognition

This talk takes a problem-oriented perspective and presents an overview of transfer-learning methods, both shallow and deep, for cross-dataset visual recognition. Specifically, itcategorisesthe cross dataset recognition into 17 problems based on a set of carefully chosen data and label attributes. Such a problem-oriented taxonomy has allowed us to examine how different transfer-learning approaches tackle each problem and how well each problem has been researched to date. This study has not only revealed the challenges in transfer learning for visual recognition but also the problems (e.g., 8 of the 17 problems) that have been scarcely studied. Also presented in this talk are two recently developed domain adaption methods, one is the joint geometrical and statistical alignment (JGSA) method and the other is adversarial nets-based unsupervised partial domain adaption.