SiamMN: Siamese modulation network for visual object tracking

Abstract

Visual object tracking methods based on Siamese network are often difficult to distinguish objects with the same semantic or similar appearance as tracking target in tracking process due to the lack of discriminating strategies for the confusing objects. We propose a visual object tracking method based on Siamese modulation network. It takes the given bounding box in the target frame and the current frame as input, and fuses these multilayer convolutional features to obtain more target appearance information of bounding box and the current frame. The feature modulator generates feature modulation vector based on the given bounding box to enhance visual appearance information of target instance in multi-layer feature of the current frame, so as to make target instance obtain higher score in response map of region proposal network, and thus realize target instancespecific tracking task. Experiments on two public benchmark datasets, OTB2015 and VOT2018, show that the proposed tracker has a competitive performance among other state-of-the art trackers.

Object Tracking Based on a Time-Varying Spatio-Temporal Regularized Correlation Filter With Aberrance Repression

“…Fu et al [32] proposed a tracking algorithm for learning discrimination and adaptive feature representation by using a hard-balanced, focus loss function and embedding an off-line training, guidance domain adaptive module in a Siamese network. Fu et al [33] proposed a visual target tracking method based on a Siamese modulation network on the basis of Resnet to extract the multi-layer fusion features of a given object in the first frame and the current frame. Meng et al [34] proposed a hierarchical correlation Siamese network for object tracking.…”