Self-supervised Video Object Segmentation by Motion Grouping
               Charig Yang   Hala Lamdouar    Erika Lu      Andrew Zisserman         Weidi Xie
                          Visual Geometry Group, University of Oxford
                        {charig,lamdouar,erika,az,weidi}@robots.ox.ac.uk
                      https://charigyang.github.io/motiongroup/
RGB Image
Our Prediction Optical Flow


Figure 1: Segmenting camouflaged animals. Motion plays a critical role in augmenting the capabi ...