ECCV 2018 Daily - Thursday

Daily Tuesday 11 7 So, what was it? Octavia explains: “ What we’re doing is exploiting temporal correlation in the dynamical data. The temporal correlation allows us to use unsupervised learning to capture the dynamics of your data using very few parameters that are invariant to linear transformations. You can model many instances of your data with just a few parameters, because we’re capturing this invariance .” This application is just for video prediction, but Octavia says their approach provides a dynamics-based encoder-decoder . It captures the dynamics and provides a way to represent video. In terms of next steps, it could be used for activity recognition, human pose prediction, video super resolution. DYAN - Octavia Camps