Paolo Andreini (DIISM, University of Siena)
Jul 5, 2018 – 9:30 AM
DIISM, Artificial Intelligence laboratory (room 201), Siena SI
In recent years, deep neural networks have become the state of the art for semantic segmentation. Nevertheless, most of the success of deep learning rests on large sets of supervised data, which are not available in many practical applications. The generation of realistic, synthetic images, together with their supervision, can greatly mitigate this problem. In this seminar, we propose some preliminary experiments and some new ideas to simultaneously generate training images and supervision for semantic segmentation.