Cross-Modality Perturbation Synergy Attack for Person Re-identification

Authors: Yunpeng Gong, Zhun Zhong, Yansong Qu, Zhiming Luo, Rongrong Ji, Min JIANG

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conducted experiments on three widely used cross-modality datasets, namely Reg DB, SYSU, and LLCM. The results not only demonstrate the effectiveness of our method but also provide insights for future improvements in the robustness of cross-modality Re ID systems.
Researcher Affiliation Academia Yunpeng Gong1, Zhun Zhong2, Yansong Qu1, Zhiming Luo1, Rongrong Ji1, and Min Jiang*1 1School of Informatics, Xiamen University 2School of Computer Science and Information Engineering, Hefei University of Technology
Pseudocode Yes Algorithm 1 Procedure of CMPS attack
Open Source Code Yes The code will be available at https://github.com/ finger-monkey/cmps__attack.
Open Datasets Yes We evaluate our proposed method on two commonly used cross-modality Re ID datasets: SYSU-MM01 [33], Reg DB [32] and LLCM [24].
Dataset Splits Yes SYSU-MM01...The testing set consists of 95 identities with two evaluation settings... Reg DB [32] is a smaller-scale dataset with 412 identities...we randomly select 206 identities (2,060 images) for training and use the remaining 206 identities (2,060 images) for testing.
Hardware Specification Yes Our experiments were conducted using three RTX 2080 Ti GPUs, each with 11GB of memory.
Software Dependencies No The paper does not explicitly list specific version numbers for software dependencies or libraries used in the experiments.
Experiment Setup Yes Here, θ represents the momentum value (set as θ = 1), and ir is derived from the previous iteration. The iteration step size is denoted by α (set as α = ϵ 12 ), where ϵ is the adversarial bound (ϵ = 8, unless otherwise specified). We set the margin ρ = 0.5 in our triplet loss.