The problem of arbitrary object tracking has traditionally been tackled by learning a model of the object’s appearance and mo- tion during the online phase. We quip a basic tracking algorithm using a Siamese network composed of fully convolutional network which incor- porates a recurrent layer at the end. The convolutional recurrent model learns a motion model and leveraging the power of Siamese network helps in achieving robust tracking. The network is trained end to end on IL- VRSC17 for similarity learning in videos.

Paper link