DarkSAM: Fooling Segment Anything Model to Segment Nothing

Authors: Ziqi Zhou, Yufei Song, Minghui Li, Shengshan Hu, Xianlong Wang, Leo Yu Zhang, Dezhong Yao, Hai Jin

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on four datasets for SAM and its two variant models demonstrate the powerful attack capability and transferability of Dark SAM. 4 Experiments
Researcher Affiliation Academia School of Computer Science and Technology, Huazhong University of Science and Technology School of Cyber Science and Engineering, Huazhong University of Science and Technology School of Software Engineering, Huazhong University of Science and Technology School of Information and Communication Technology, Griffith University {zhouziqi,yufei17,minghuili,hushengshan,wxl99,dyao,hjin}@hust.edu.cn leo.zhang@griffith.edu.au
Pseudocode Yes Algorithm 1 Dark SAM
Open Source Code Yes Our codes are available at: https://github.com/CGCL-codes/Dark SAM.
Open Datasets Yes We evaluate our method using four public segmentation datasets: ADE20K [43], MS-COCO [23], CITYSCAPES [7], and SA-1B [19].
Dataset Splits No For each dataset, we randomly select 100 images for UAP generation and 2,000 images for testing purposes.
Hardware Specification Yes Experiments are conducted on a server running a 64-bit Ubuntu 20.04.1 system with an Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz processor, 125GB memory, and two Nvidia Ge Force RTX 3090 GPUs, each with 24GB memory.
Software Dependencies Yes The experiments are performed using the Python language and Py Torch library version 2.1.0.
Experiment Setup Yes For our experiments, we adjust the hyperparameters k, τ, λ, and µ to 10, 1, 0.1, and 0.01, respectively, and set the batch size to 1.