Visual Tracking Jointly With Online and Offline Learning

While recent approaches based on offline learning perform well in balancing the accuracy and speed of tracking, it is still non-trivial to accommodate a pre-trained model to an unseen target. In fact, online learning, which requires to capture specified target characteristics from one shot, is a unique attribute of single object tracking. How to favorably bridge the gap between the offline and online learning is of importance for tracking any unseen targets. In this work, in order to obtain sequence-specific information, we propose an online lightweight network consisting of feature adapting layer and ridge regression layer. Its key innovation is to interpret the ridge regression as one layer of the network. Furthermore, we integrate cross-similarity into the Siamese network and train it offline in an end-to-end manner to acquire the fine-grained local pattern of the target object. Through our effective fusion scheme for the offline and online procedures, our method can achieve considerable improvements on prevalent benchmarks.

Abstract licence: CC BY-NC-ND

40 more references, whose wording the paper’s licence keeps private.

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.