返回
Siamese network with transformer and saliency encoder for object tracking
DOI:10.1007/s10489-022-03352-3.png)
摘要
En 中文
Siamese network-based tracking algorithms are highly prone to lose semantic information as well as detailed features of objects when applying correlation operations, and they lack global modeling capabilities that make it difficult to track objects in multiple complex scenarios. To address the above problems, this paper proposes a feature repair strategy combining Transformer and saliency encoder to repair the feature loss from correlation operation. We first add a saliency encoder network branch that is parallel with Siamese network to provide more detailed features and semantic information for the regression and classification to reduce the interference from invalid objects. Second, we fuse the correlation response graph with the encoded saliency features and use the encoding part of Transformer to enhance the nonlinear ability of the fused feature graph to capture global contextual information. The integrated and enhanced feature map can effectively optimize the classification and localization capabilities of our algorithm. Finally, the DIoU loss function is used to continuously optimize the generation of bounding boxes during training. The algorithm proposed in this paper achieves advanced performance in experiments on five publicly available datasets, namely, GOT-10k, LaSOT, UAV123, DTB70, and TrackingNet.
Keyword:
Siamese network
Correlation
Transformer
Saliency encoder
DIoU
期刊
IF:
3.5
论文数:
7.6K
被引数:
1.7W
机构
引用论文
Cobalt oxides nanoparticles supported on nitrogen-doped carbon nanotubes as high-efficiency cathode catalysts for microbial fuel cells负载在氮掺杂碳纳米管上的钴氧化物纳米颗粒作为微生物燃料电池的高效阴极催化剂
A novel flake-ball-like magnetic Fe3O4/γ-MnO2 meso-porous nano-composite: Adsorption of fluorinion and effect of water chemistry
Chemosphere
IF0
Metabolic profiles of three fiber types of skeletal muscle in guinea pigs and rabbits
Biochemistry
IF0

