no code implementations • 8 May 2019 • Hieu Le, Bento Gonçalves, Dimitris Samaras, Heather Lynch
This segmentation network is trained with a specific loss function, based on the average activation, to effectively learn from the data with the weakly-annotated labels.