LT1st / awesome-Adversarial-Attack-for-Visual-Object-Tracking

A paper collection of Adversarial Attack for Visual Object Tracking.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

awesome-Adversarial-Attack-for-Visual-Object-Tracking

A paper collection of Adversarial Attack for Visual Object Tracking.

论文 相关痛点及方法(分点简要总结) 方法 时间 期刊
Tracklet-Switch Adversarial Attack against Pedestrian Multi-Object Tracking Trackers MOT 的关联算法对跟踪过程中的错误具有鲁棒性;通过扰动很少的帧来跟踪后续帧中的目标 第一个针对行人 MOT 跟踪器的对抗性攻击
Efficient Adversarial Attacks for Visual Object Tracking l drift loss combined with the embedded feature loss eccv_2020
IoU Attack: Towards Temporally Coherent Black-Box Adversarial Attack for Visual Object Tracking IoU 攻击,它根据当前帧和历史帧的预测 IoU 分数顺序生成扰动。通过降低 IoU 分数 降低了时间相干边界框(即对象运动)的准确性 cvpr 21
Only Once Attack: Fooling the Tracker With Adversarial Template 仅针对初始帧进行攻击;攻击次数少 基于最小分数和基于最小 IoU 的损失函数
AdvBokeh: Learning to Adversarially Defocus Blur 深度引导的虚化合成网络 (DebsNet),能够灵活地合成、重新聚焦和调整图像的虚化程度 eccv_2020
SPARK: Spatial-Aware Online Incremental Attack Against Visual Tracking 跟踪对象的移动轨迹
Cooling-Shrinking Attack: Blinding the Tracker with Imperceptible Noises 冷却热图上目标存在的热点区域,并迫使预测的边界框收缩,使跟踪目标对跟踪器不可见 冷却热图上目标存在的热点区域,并迫使预测的边界框收缩,使跟踪目标对跟踪器不可见
Physical adversarial textures that fool visual object tracking. 纹理以数字或印刷海报的形式显示在物理世界中时,会导致视觉对象跟踪系统变得混乱 (EOT)算法来生成在不同条件下成像时欺骗跟踪模型的物理对手,但我们比较了不同调节变量(包括视点、照明和外观)的影响,以找到具有高结果的实用攻击设置对抗强度和收敛速度
One-Shot Adversarial Attacks on Visual Tracking With Dual Attention 初始帧中的目标块上添加轻微的扰动 CVPR 22
Universal Physical Camouflage Attacks on Object Detectors UPC 通过联合欺骗区域提议网络;对非刚性或非平面物体有效,我们引入了一组用于模仿可变形属性的变换;第一个标准化虚拟数据库 AttackScenes
SPARK: Spatial-Aware Online Incremental Attack Against Visual Tracking 跟踪跟踪对象的移动轨迹;在线生成;1)很难生成可以跨帧传输的难以察觉的扰动,2)实时跟踪器要求攻击满足一定的效率水平。 使用 30 作为攻击间隔,因为视频通常为 30 fps,这样的设置自然利用了第 29 帧和第 30 帧之间的潜在延迟 ECCV 2020

Paper with code

  1. FairMOT-attack

FairMOT-attack

  1. IoUattack

IoUattack https://arxiv.org/pdf/2103.14938.pdf

  1. jadena

"Can You Spot the Chameleon? Adversarially Camouflaging Images from Co-Salient Object Detection" in CVPR 2022

https://github.com/tsingqguo/jadena

4.CSA

CVPR2020 Paper Cooling-Shrinking Attack: Blinding the tracker with imperceptible noises

https://github.com/MasterBin-IIAU/CSA

Paper without code

About

A paper collection of Adversarial Attack for Visual Object Tracking.

License:Apache License 2.0