Zhenbang Li

ORCID: 0000-0002-3872-6777
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Video Surveillance and Tracking Methods
  • Human Pose and Action Recognition
  • Anomaly Detection Techniques and Applications
  • Adversarial Robustness in Machine Learning
  • Railway Engineering and Dynamics
  • Face recognition and analysis
  • Fire Detection and Safety Systems
  • Tribology and Lubrication Engineering
  • Gear and Bearing Dynamics Analysis

Shanghai Institute of Technology
2023

Chinese Academy of Sciences
2020-2021

Beijing Academy of Artificial Intelligence
2020-2021

Institute of Automation
2020-2021

University of Chinese Academy of Sciences
2020-2021

Siamese trackers are shown to be vulnerable adversarial attacks recently. However, the existing attack methods craft perturbations for each video independently, which comes at a non-negligible computational cost. In this paper, we show existence of universal that can enable targeted attack, e.g., forcing tracker follow ground-truth trajectory with specified offsets, video-agnostic and free from inference in network. Specifically, by adding translucent perturbation template image <italic...

10.1109/tcsvt.2021.3120479 article EN IEEE Transactions on Circuits and Systems for Video Technology 2021-10-16

In this letter, we show that the challenging model adaptation task in visual object tracking can be handled by simply manipulating pixels of template image Siamese networks. For a target is not included offline training set, slight modification will improve prediction result trained network. The popular adversarial example generation methods used to perform pixel manipulation for adaptation. Different from current update methods, which aim combine features previous frames, focus on initial...

10.1109/lsp.2020.3025406 article EN IEEE Signal Processing Letters 2020-01-01

While siamese networks have demonstrated the significant improvement on object tracking performances, how to utilize temporal information in trackers has not been widely studied yet. In this paper, we introduce a novel architecture equipped with aggregation module, which improves per-frame features by aggregating from adjacent frames. This fusion strategy enables handle poor appearance like motion blur, occlusion, etc. Furthermore, incorporate adversarial dropout module network for computing...

10.1109/icip40778.2020.9191165 article EN 2022 IEEE International Conference on Image Processing (ICIP) 2020-09-30

As rail vehicles cover more mileage and run at higher speeds, the problem of abnormal wear between wheel becomes increasingly prominent. Identifying polygons is a crucial aspect tackling this issue. In study, acceleration signals are collected using rotation angle as reference instead time, signal processing carried out spatial spectrum Fourier transform. The results demonstrate that transform based on can accurately identify frequency polygon distribution. This method solves order...

10.1117/12.2689604 article EN 2023-09-07

Siamese trackers are shown to be vulnerable adversarial attacks recently. However, the existing attack methods craft perturbations for each video independently, which comes at a non-negligible computational cost. In this paper, we show existence of universal that can enable targeted attack, e.g., forcing tracker follow ground-truth trajectory with specified offsets, video-agnostic and free from inference in network. Specifically, by adding imperceptible perturbation template image fake...

10.48550/arxiv.2105.02480 preprint EN cc-by arXiv (Cornell University) 2021-01-01
Coming Soon ...