Computer Vision News - July 2022

BEST OF CVPR 8 CVPR Poster Presentation knowledge and the new annotations. “We started with two things in mind,” Dario reveals. “We wanted something that could be used in the real world because acquiring annotation, which is usually very expensive, is a challenge we all face. We also wanted to create a model that could incrementally extend its internal knowledge over time.” Fabio tells us that finding a jumping-off point for developing the model was the most challenging part: “We tried lots of different things in the beginning. You won’t read about those in the paper because they’re failed experiments! The tough part was finding something that might work, even if not that well, but just as a starting point to develop our method.” The team mixed two well-known problems: incremental or continual learning and weakly supervised segmentation . Continual learning is a fundamental idea in machine learning, where you want to extend your model over time. It has been investigated in image classification and recently in semantic segmentation. Weakly supervised semantic segmentation is also This work aims to incrementally extend a semantic segmentation model with new classes over time without using expensive per pixel annotation . Instead, the model relies on image-level labels, which are cheap and can be found easily online. The goal is to exploit the model to extract the boundaries of the new classes by itself and then learn the new classes with this INCREMENTAL LEARNING IN SEMANTIC SEGMENTATION FROM IMAGE LABELS Fabio Cermelli (left) and Dario Fontanel (right) are third-year PhD students at Politecnico di Torino in Italy under the supervision of Barbara Caputo. They speak to us ahead of their poster presentation.