Structural pixel-wise target attention for robust object tracking

Zhang, Huanlong; Cheng, Liyun; Zhang, Jianwei; Huang, Wanwei; Liu, Xiulei; Yu, Junyang

doi:10.1016/j.dsp.2021.103139

Cited by 10 publications

(3 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The attention mechanism (Mnih et al, 2014) in a neural network helps to allocate the computing resources to more important tasks and effectively deal with the problems with overload data information. As shown in Figure 3, the core idea of attention mechanism is to find the correlation of the original data and then highlight their important features, such as channel attention (Wang et al, 2020), pixel attention (Zhang et al, 2021), etc. Without loss of generality, consider an input of v i for a given data set v ¼ ðv 1 ; v 2 ; .…”

Section: Attention Mechanismmentioning

confidence: 99%

An encoder-decoder model with embedded attention-mechanism for efficient meshfree prediction of slope failure

Chen,

Wang,

Deng

et al. 2023

International Journal of Damage Mechanics

View full text Add to dashboard Cite

The particle-based meshfree methods provide an effective means for large deformation simulation of the slope failure. Despite the advances of various efficient meshfree algorithmic developments, the computational efficiency still limits the application of meshfree methods for practical problems. This study aims at accelerating the meshfree prediction of the slope failure through introducing an encoder-decoder model, which is particularly enhanced by the attention-mechanism. The encoder-decoder model is designed to capture the long sequence character of meshfree slope failure analysis. The discretization flexibility of meshfree methods offers an easy match between the meshfree particles and machine learning samples and thus the resulting surrogate model for meshfree slope failure prediction has a quite wide applicability. In the meantime, the embedding of the attention-mechanism into the encoder-decoder neural network not only enables a significant reduction of the number of meshfree model parameters, but also maintains the key features of meshfree simulation and effectively alleviates the information dilution issue. It is shown that the proposed encoder-decoder model with embedded attention mechanism gives a more favorable prediction on the meshfree slope failure simulation in comparison to the general encoder-decoder formalism.

show abstract

Section: Attention Mechanismmentioning

confidence: 99%

An encoder-decoder model with embedded attention-mechanism for efficient meshfree prediction of slope failure

Chen,

Wang,

Deng

et al. 2023

International Journal of Damage Mechanics

View full text Add to dashboard Cite

show abstract

“…Video object tracking, which refers to continuously tracking the state of an object in subsequent frame sequences by using the initial position and scale information of the object, is the basis for high-level visual tasks such as visual inspection, visual navigation, and visual servo (Nousi et al, 2020 ; Wang et al, 2020 ; Karakostas et al, 2021 ; Sun et al, 2021 ). In engineering practice, interference such as changes in the posture and scale of the object, noise interference, background occlusion, or variation of light conditions may lead to tracking failure, so object tracking remains a challenging task (Zhang et al, 2020 ; Zhang H. et al, 2021 ; Liu et al, 2022 ).…”

Section: Introductionmentioning

confidence: 99%

A Siamese tracker with “dynamic–static” dual-template fusion and dynamic template adaptive update

et al. 2023

View full text Add to dashboard Cite

In recent years, visual tracking algorithms based on Siamese networks have attracted attention for their desirable balance between speed and accuracy. The performance of such tracking methods relies heavily on target templates. Static templates cannot cope with the adverse effects of target appearance change. The dynamic template method, with a template update mechanism, can adapt to the change in target appearance well, but it also causes new problems, which may lead the template to be polluted by noise. Based on the DaSiamRPN and UpdateNet template update networks, a Siamese tracker with “dynamic–static” dual-template fusion and dynamic template adaptive update is proposed in this paper. The new method combines a static template and a dynamic template that is updated in real time for object tracking. An adaptive update strategy was adopted when updating the dynamic template, which can not only help adjust to the changes in the object appearance, but also suppress the adverse effects of noise interference and contamination of the template. The experimental results showed that the robustness and EAO of the proposed method were 23% and 9.0% higher than those of the basic algorithm on the VOT2016 dataset, respectively, and that the precision and success were increased by 0.8 and 0.4% on the OTB100 dataset, respectively. The most comprehensive real-time tracking performance was obtained for the above two large public datasets.

show abstract

“…Video object tracking, which refers to continuously tracking the state of the object in subsequent frame sequences by using the initial position and scale information of the object, is the basis for high-level visual tasks such as visual inspection, visual navigation and visual servo [1][2][3][4]. In engineering practice, interference such as changes in the posture and scale of the object, noise interference, background occlusion or variation of light condition may lead to tracking failure, so object tracking is still a challenging task [5][6][7]. Object tracking methods can be roughly divided into generative methods and discriminative methods [8][9].…”

Section: Introductionmentioning

confidence: 99%

Siamese Tracker with “Dynamic-static” Dual-template Fusion and Dynamic Template Adaptive Update

Sun

Wang

Man

et al. 2022

Preprint

View full text Add to dashboard Cite

In recent years, most visual tracking algorithms based on Siamese network have attracted much attention for its desirable balance between speed and accuracy. The performance of such tracking methods relies heavily on target templates, and there will appear some problems in the application whether dynamic or static templates are used. Based on DaSiamRPN and UpdateNet template update network, a Siamese tracker with “dynamic-static” dual-template fusion and dynamic template adaptive update is proposed in this paper. The new method combines a static template and a dynamic template that is updated in real time for object tracking. An adaptive update strategy is adopted when updating the dynamic template, which can not only help adjust to the changes in the object appearance but also suppress the adverse effects of noise interfering and contaminating the template. Experimental results show that the robustness and EAO of the proposed method are 23% and 9.0% higher than the basic algorithm on the VOT2016 dataset respectively and that the precision and success are increased by 0.8% and 0.4% on the OTB100 dataset respectively. The best comprehensive performance is obtained on the above two large public datasets, and the anti-interference capability is significantly improved.

show abstract

Structural pixel-wise target attention for robust object tracking

Cited by 10 publications

References 18 publications

An encoder-decoder model with embedded attention-mechanism for efficient meshfree prediction of slope failure

An encoder-decoder model with embedded attention-mechanism for efficient meshfree prediction of slope failure

A Siamese tracker with “dynamic–static” dual-template fusion and dynamic template adaptive update

Siamese Tracker with “Dynamic-static” Dual-template Fusion and Dynamic Template Adaptive Update

Contact Info

Product

Resources

About