Computer Vision News - September 2022

22 Congrats, Doctor! available, or the data annotation is costly and require expert knowledge. To mitigate this issue, learning with limited data has gained considerable attention. In this work, two approaches to handle the availability of limited annotated data, including open- set recognition and domain adaptation, are studied. The classification problem is investigated under an open-world assumption to handle unpredictable categories in a long-tail distribution dataset. For the open-set recognition, a representative- discriminative multi-task learning framework is presented, as illustrated in Figure 1. The proposed framework operates among three spaces, including 1) the transformation from the raw image space to the embedding feature space, and 2) the transformation from the embedding feature space to a so-called abundance space. The first transformation creates a more discernible input to be fed into the subsequent open-set learning network. The second transformation provides a finer-scale representation which is useful for the unknown detection part. Hence, it exploits both the representative and discriminative aspects of data in order to best characterize the differences between known and unknown classes. Experiments on multiple satellite benchmarks and RGB image datasets demonstrate significant improvement over state-of-the-art open- set recognition algorithms. In addition, the generalization capability of a recognition system is studied where the goal is to build a model which has a reasonable performance across diverse Razieh Kaviani Baghbaderani recently completed her PhD at the University of Tennessee- Knoxville. Her research mainly focused on learning with limited labeled data for image and video understanding. During her PhD, she also worked as a research intern at Samsung Semiconductor Inc and American Family Insurance. Currently, she pursues her career as a research scientist in the multimedia systems team at Samsung Semiconductor Inc. - Congrats, Doctor Razieh! Despite the great progress of deep learning in many computer vision tasks, it often demands a large-scale and carefully annotated dataset. However, it is not practical in many real-world applications where only a few examples may be