SJDL-Vehicle: Semi-supervised Joint Defogging Learning for Foggy Vehicle Re-identification

Authors: Wei-Ting Chen, I-Hsiang Chen, Chih-Yuan Yeh, Hao-Hsiang Yang, Jian-Jiun Ding, Sy-Yen Kuo347-355

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that the proposed method is effective and outperforms other existing vehicle Re ID methods in the foggy weather.
Researcher Affiliation Collaboration Wei-Ting Chen1,3 , I-Hsiang Chen2 , Chih-Yuan Yeh2, Hao-Hsiang Yang2, Jian-Jiun Ding2, and Sy-Yen Kuo2 1Graduate Institute of Electronics Engineering, National Taiwan University, Taiwan 2Department of Electrical Engineering, National Taiwan University, Taiwan 3ASUS Intelligent Cloud Services, Taiwan {f05943089, f09921058, f09921063, r05921014, jjding, sykuo}@ntu.edu.tw
Pseudocode No The paper describes the proposed architecture and processes in prose but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes The code and dataset are available in https://github.com/Cihsaing/SJDLFoggy-Vehicle-Re-Identification--AAAI2022.
Open Datasets Yes Due to the lack of a dataset specialized for vehicle Re ID in the foggy weather, we construct a dataset called FVRID which consists of real-world and synthetic foggy images to train and evaluate the performance. The code and dataset are available in https://github.com/Cihsaing/SJDLFoggy-Vehicle-Re-Identification--AAAI2022.
Dataset Splits No The paper provides details for 'Train', 'Probe', and 'Gallery' sets, but does not explicitly mention or detail a separate 'validation' dataset split for hyperparameter tuning or early stopping during training. 'Probe' is used for evaluation/testing.
Hardware Specification Yes The network is trained on an Nvidia Tesla V100 GPU for 20 hours and we implement it on the Pytorch platform.
Software Dependencies No The paper states 'we implement it on the Pytorch platform' but does not specify a version number for PyTorch or any other software dependencies.
Experiment Setup Yes The input image is resized to 384 x 384 and the training batch size Q is set to 36. We apply horizontal flip and random crop to prevent the overfitting problem due to the limited number of training data. We train models for 120 epochs with a warm-up strategy. The initial learning rate is 1.09 x 10^-5, which increases to 10^-4 after the 10th epoch. The Adam optimizer is adopted to optimize the model with a decay rate of 0.6. The hyper-parameters λ1, λ2, λ3, and λ4 are set as to 1, 10^-5, 10^-5, and 300.