Computer Science News Archive
Carlo Innamorati wins the Best Student Paper Honourable Mention Award
The paper, Learning on the Edge: Explicit Boundary Handling in CNNs, proposes an innovative solution as an alternative to commonly used padding strategies in convolutions for Convolutional Neural Networks. The main insight is represented by the fact that standard padding strategies introduce a bias that is propagated through the entire pipeline. The alternative presented in the paper allows networks to learn to alleviate this bias on their own. The method proved to be more effective than established strategies on a variety of image processing tasks such as De-bayering and Colorization. For more details, please visit http://geometry.cs.ucl.ac.uk/projects/2018/learning-edge/.
Carlo Innamorati has completed an MSc in Computer Graphics and is now in his third year of PhD here at UCL, Supervised by Niloy J. Mitra and Tobias Ritschel. The project is funded by Marie Sk?odowska-Curie grant agreement, the ERC Starting Grant SmartGeometry, and the UK Engineering and Physical Sciences Research Council.