Transferable Adversarial Attacks for Image and Video Object Detection

Authors: Xingxing Wei, Siyuan Liang, Ning Chen, Xiaochun Cao

IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on PASCAL VOC and Image Net VID datasets show that our method efficiently generates image and video adversarial examples, and more importantly, these adversarial examples have better transferability, therefore being able to simultaneously attack two kinds of representative object detection models: proposal based models like Faster RCNN and regression based models like SSD.
Researcher Affiliation Academia 1Dept. of Comp. Sci. & Tech., Institute for Artificial Intelligence, State Key Lab for Intell. Tech. & Sys., THBI Lab, Tsinghua University 2Institute of Information Engineering, Chinese Academy of Sciences
Pseudocode No The paper describes the methodology using text and mathematical equations but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement regarding the release of their own source code or a direct link to a code repository for the methodology described.
Open Datasets Yes For image detection, we use the training dataset of PASCAL VOC 2007 with totally 5011 images to train the adversarial generator. ... For video detection, we use Image Net VID dataset. ... http://bvisionweb1.cs.unc.edu/ILSVRC2017/download-videos1p39.php
Dataset Splits No The paper states using PASCAL VOC 2007 training dataset for training and PASCAL VOC 2007 testing set for testing, but does not explicitly mention or detail a separate validation dataset split.
Hardware Specification No The paper does not provide specific details on the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions using the Adam solver and refers to external implementations of Faster-RCNN and SSD300 with links, but it does not specify version numbers for any software dependencies or libraries crucial for replication.
Experiment Setup Yes We set α = 0.05, β = 1. For ϵ, we set 1 10 4 and 2 10 4 for the selected two layers, respectively. To optimize our networks under Eq.(5), we follow the standard approach from [Isola et al., 2017] and apply the Adam solver [Kingma and Ba, 2014]. The best weights are obtained after 6 epochs.