Computer Vision News - October 2018

Daily Thursday Poster Presentations 25 Pauline Luc presenting her poster “Predicting Future Instance Segmentation by Forecasting Convolutional Features”. She told us that “ predicting abstract convolutional features drives futures instant segmentation .” Advisors Jakob Verbeek added: “ Most work on future prediction is based at raw RGB level. Our work instead predicts futures at highly semantic level, which is directly useful for diverse application .” DmytroMishkin explaining his work: “Repeatability Is Not Enough: Learning Affine Regions via Discriminability”