Learning Location Sensitive Embeddings for Video Object Segmentation

This paper proposes a method to learn features which are aware of the object locations. When combined with the translation invariant convolution features, it can differentiate similar objects at different locations.