Full-Distance Evasion of Pedestrian Detectors in the Physical World

Authors: Zhi Cheng, Zhanhao Hu, Yuqiu Liu, Jianmin Li, Hang Su, Xiaolin Hu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our physical world experiments demonstrate the effectiveness of our FDA patterns across various detection models like YOLOv5, Deformable-DETR, and Mask RCNN.
Researcher Affiliation Academia Zhi Cheng1, Zhanhao Hu2, Yuqiu Liu3, Jianmin Li1, Hang Su1, Xiaolin Hu1* 1Department of Computer Science and Technology, Tsinghua University, Beijing, China 2Department of Electrical Engineering and Computer Sciences, UC Berkeley 3Department of Technology, Beijing Forestry University, Beijing, China
Pseudocode No The paper describes the methodology and various modules (DIC, MFO) but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes Codes available at https://github.com/zhicheng2T0/Full-Distance Attack.git
Open Datasets Yes To optimize the FDA patterns in the digital world, we created a pedestrian dataset and a background dataset. 1100 pedestrian images were extracted from existing datasets (INRIA [3], Penn Fudan [36] and COCO [24]).
Dataset Splits Yes To form a distant image dataset to train the DIC (Figure 4 (a)), we printed 45 training images and 9 testing images onto papers, collected photos of all printed images at 7 distances (4m, 8m, 14m, 20m, 26m, 34m, 40m) in 5 days and removed ones with noises (e.g. reflections and shadows).
Hardware Specification No The paper mentions specific smartphone cameras used for capturing test images ('back camera of Xiaomi-CIVI smart phone', 'Huawei-Nova-11-SE and OPPO-A9 smart phones'), but it does not specify any hardware details (like GPU or CPU models, memory) used for running the computational experiments or training models.
Software Dependencies No The paper references various detection models like YOLOv5, Deformable-DETR, and Mask RCNN, but it does not provide specific version numbers for these or other software dependencies like programming languages or libraries.
Experiment Setup No The paper states 'For optimization details, we used configurations of Adv-Tshirt [42] and TCA [17] for patch and clothing experiments respectively,' which defers specific hyperparameters to external works rather than detailing them in the main text.