Attention Disturbance and Dual-Path Constraint Network for Occluded Person Re-identification

Authors: Jiaer Xia, Lei Tan, Pingyang Dai, Mingbo Zhao, Yongjian Wu, Liujuan Cao

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

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experimental evaluations conducted on person re-ID benchmarks demonstrate the superiority of ADP over stateof-the-art methods.
Researcher Affiliation Collaboration Jiaer Xia1*, Lei Tan1*, Pingyang Dai1 , Mingbo Zhao2, Yongjian Wu3, Liujuan Cao1 1Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, China 2Donghua University, Shanghai, China 3Tencent Youtu Lab, Shanghai, China {xiajiaer, tanlei}@stu.xmu.edu.cn, pydai@xmu.edu.cn, mzhao4@dhu.edu.cn, littlekenwu@tencent.com, caoliujuan@xmu.edu.cn
Pseudocode No The paper describes the proposed method using text and mathematical equations, but it does not include pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not contain any explicit statements or links indicating that the source code for the described methodology is publicly available.
Open Datasets Yes To validate the effectiveness of our proposed method, we perform extensive experiments on publicly available Re-ID datasets, including both occluded (Miao et al. 2019; Zhuo et al. 2018) and holistic (Zheng et al. 2015; Zheng, Zheng, and Yang 2017; Ristani et al. 2016) datasets. Occluded-Duke (Miao et al. 2019) is a large-scale dataset selected from the Duke MTMC for occluded person reidentification. Occluded-REID (Zhuo et al. 2018) is an occluded person dataset captured by mobile cameras. Market-1501 (Zheng et al. 2015) is a widely-used holistic Re-ID dataset captured from 6 cameras. Duke MTMC-re ID (Zheng, Zheng, and Yang 2017; Ristani et al. 2016) contains 36,441 images of 1,812 persons captured by 8 cameras
Dataset Splits No The paper specifies training sets, query sets, and gallery sets for evaluation, but it does not explicitly define or mention a separate 'validation' dataset split for hyperparameter tuning or early stopping.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU models, CPU specifications, memory).
Software Dependencies No The paper mentions using 'ViT ... pre-trained on Image Net as our backbone' but does not specify software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions).
Experiment Setup Yes The input images are resized to 256 128 and augmented by commonly used random horizontal flipping, padding and random cropping. During the training phase, the batch size is set to 64 with 16 identities. We utilize the SGD as the optimizer, with the initial learning rate of 0.004 and a cosine learning rate decay. The margin of each triplet loss is set to 0.3. The hyper-parameter m and s in eq.(9) are set to 0.3 and 30, respectively, while the λ in eq.(12) is 0.1.