Computer Vision News - January 2023

12 Trans2k, Best Paper BMVC The team must be doing something right because scooping the Best Paper Award at an international conference is no mean feat. Can Alan, Žiga, and Matej offer any insights into what they think captured the judges’ attention? “ Maybe they realized that the paper is useful for a wider audience, ” Alan ponders. “ It could be useful in the industry – there are many people in the glass industry, for example. Also, it’s presented for tracking transparent objects, but it could be applied to other fields, such as object detection. I think the paper shows how to extend the training dataset for other problems . ” Matej agrees: “ When we started the challenges, the subfields of computer vision were separated methodologically, ” Matej recalls. “ You could draw from other fields, but not directly. We’re always drawing from object detection, feature matching, and geometry when we’re designing tracking algorithms, but what changed was deep learning . Deep learning came along, and the computer vision field started to converge methodologically. That allowed us to disseminate results faster across the subfields. We’re now drawing more strongly from subfields such as detection, optical flow, and motion estimation and integrating those into new algorithms based on deep learning . ”