ECCV 2016 Daily - Tuesday

Video Segmentation researchers working on video analysis. The workshop covered a number of hot topics, which relate to video segmentation and its role in learning. Among the others, both Jitendra Malik and Vittorio Ferrari stressed the importance of large amounts of data for state-of-the-art deep networks. As Vittorio showed, video segmentation is already a practical path to automatic data labelling in many respects. The presentations of Thomas Brox , Michael Black and Michal Irani related to the important topic of video representation, in relation to tasks such as segmentation, the estimation of optical flow and the recognition of activities. This kept our panel in a lively discussion, showing that there is still some way to go from the potentially well understood topic of optical flow to the work-in-progress action recognition arena. The overflowing room during the entire workshop was indeed evidence of how hot video analysis is today. Fabio Galasso heads the Sensors Department at OSRAM in Germany A word from the organizers It is an exciting time for computer vision and even more so for al l Workshop 7 ECCV Daily: Tuesday The first invited talk of the workshop was given by Jitendra Malik , of the University of California - Berkeley . Jitendra started with a brief history of motion analysis , before computer vision came into the scene, starting from important work on movement by Max Wertheimer , one of the founders of the Gestalt school. In the second part of his presentation, Jitendra reviewed the contribution of computer vision to motion analysis , starting from reconstruction and recognition, where motion cues have to be detected and classified. Jitendra says that we should develop models which are unsupervised or self-supervised , taking inspiration from infant development , in particular the “Six Lessons from Babies” of Linda Smith and Michael Gasser: • Be multi-modal • Be incremental • Be physical • Explore • Be social • Use language Jitendra concluded showing this hilarious video , which is intriguing because it highlights how segmenting individual people might conflict with color grouping cues and with common movement; it illustrates the fundamental dilemma facing us: the low-level cues (color and movement) conflicts with the high-level cues of what does the person look like.