Visual Recognition
The goal of this project is to create efficient visual recognition systems that scale with the vast number of categories. This include recognizing objects in a scene, image classification, and image retrieval among others.Currently, the laboratory is interested in developing visual recognition systems that can learn in a more human-like scenario. The team currently investigates solutions to the open-set recognition problem. In this scenario, all the visual concepts to recognize are not well defined. This means that the recognition system can discover visual categories that system never learned during training time. This is a challenging but exciting problem, since it opens the door to novel solutions that enable systems to learn perpetually.
Publications:
1. V. Fragoso, W. Scheirer, J. Hespanha, M. Turk. One-class Slab Support Vector Machine. IAPR ICPR. 2016.
2. C. Torres, V. Fragoso, S. Hammond, J. Fried, B.S. Manjunath. Eye-CU: Sleep Pose Classification for Healthcare using Multimodal Multiview Data. IEEE WACV. 2016.