Deep Low-Resolution Person Re-Identification

Authors: Jiening Jiao, Wei-Shi Zheng, Ancong Wu, Xiatian Zhu, Shaogang Gong

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive evaluations show the advantages of our method over related state-of-the-art re-id and super-resolution methods on cross-resolution re-id benchmarks. We extensively performed comparative evaluations to show the superiority of our SING approach over related state-of-the-art re-id and image SR methods on four person re-id benchmarks CAVIAR (Cheng et al. 2011), VIPe R (Gray and Tao 2008), CUHK03 (Li et al. 2014), and SYSU (Chen et al. 2017a). Experiments Datasets. We performed evaluations on three simulated and one genuine LR person re-id datasets (Fig. 4).
Researcher Affiliation Academia Jiening Jiao,1,2 Wei-Shi Zheng,3,4 Ancong Wu,1 Xiatian Zhu,5 Shaogang Gong5 1 School of Electronics and Information Technology, Sun Yat-sen University, China 2 Collaborative Innovation Center of High Performance Computing, NUDT, China 3 School of Data and Computer Science, Sun Yat-sen University, China 4 Key Laboratory of Machine Intelligence and Advanced Computing (Sun Yat-sen University), Ministry of Education, China 5 School of Engineering and Computer Science, Queen Mary University of London, UK jiaojn@mail2.sysu.edu.cn, wszheng@ieee.org, wuancong@mail2.sysu.edu.cn, xiatian.zhu@qmul.ac.uk, s.gong@qmul.ac.uk
Pseudocode No The paper describes the network architecture but does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide any statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets Yes Datasets. We performed evaluations on three simulated and one genuine LR person re-id datasets (Fig. 4). (1) MLR-VIPe R was constructed from the VIPe R (Gray and Tao 2008) dataset. (2) MLR-SYSU is based on the SYSU dataset (Chen et al. 2017a). (3) MLR-CUHK03 was built from the CUHK03 (Li et al. 2014) dataset. (4) CAVIAR is a genuine LR person re-id dataset (Cheng et al. 2011).
Dataset Splits Yes All datasets except MLRCUHK03 were randomly divided into two halves, one for training and one for testing. That is, there are p = 25, 316 and 251 persons in the testing set of CAVIAR, MLR-VIPe R and MLR-SYSU, respectively. Following (Xiao et al. 2016), we utilised the benchmarking 1,367/100 training/test identity split. We repeated 10 times of the above random data split. The scaling parameter σ in Eq. (8) was set by cross validation on the validation set.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU models, CPU types, memory) used for running the experiments.
Software Dependencies No The paper mentions the use of deep CNN models (SRCNN, DGD network) but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup Yes In implementation, we upscale the LR images to an appropriate size (160 72 in our experiments) by bicubic interpolation as (Dong et al. 2016). We initialised the SR and Re ID sub-networks by SRCNN (Dong et al. 2016) pre-trained on Image Net-1K and DGD (Xiao et al. 2016) pre-trained on the training data of Market-1501 (Zheng et al. 2015), respectively. The scaling parameter σ in Eq. (8) was set by cross validation on the validation set. We set the balance coefficient α = 1 (Eq. (5)) which assumes equal importance between image SR and re-id feature learning.