ECE Research Seminar-Deep learning methods for instance segmentation

In this talk I will present deep learning methods for inferring sequential visual content. I will focus on instance segmentation, which is the problem of detecting and delineating each distinct object of interest appearing in an image. Traditional instance segmentation approaches consist of ensembles of modules that are trained independently of each other, thus missing opportunities for joint learning. Here I will present an instance segmentation paradigm consisting in an end-to-end method that learns how to segment instances sequentially. The model is based on a recurrent neural network that sequentially finds objects and their segmentations one at a time. This net is provided with a spatial memory that keeps track of what pixels have been explained and allows occlusion handling. In order to train the model we designed a principled loss function that accurately represents the properties of the instance segmentation problem. As we will see, this method outperforms recent approaches on multiple person segmentation, and leaf counting. I will conclude presenting a short overview of other instance segmentation approaches published recently.

Bio: Bernardino Romera-Paredes is a research scientist in DeepMind. He was a postdoctoral research fellow in the Torr Vision Group at the University of Oxford. Before that, he received his PhD degree from University College London in 2014, supervised by Prof. Massimiliano Pontil and Prof. Nadia Berthouze. He has published in top-tier machine learning conferences such as NIPS, ICML, ICCV, and AISTATS. During his PhD he interned at Microsoft Research, Redmond. His research focuses on multitask and transfer learning methods applied to Computer Vision tasks such as object recognition and segmentation, and emotion recognition.

Presented by

Key learning outcomes