Cross-Modality Earth Mover’s Distance for Visible Thermal Person Re-identification
Authors: Yongguo Ling, Zhun Zhong, Zhiming Luo, Fengxiang Yang, Donglin Cao, Yaojin Lin, Shaozi Li, Nicu Sebe
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show the benefits of the proposed CM-EMD and its auxiliary techniques (CMDL and MGS). Our method achieves state-of-the-art performance on two VT-Re ID benchmarks. |
| Researcher Affiliation | Academia | 1Department of Artificial Intelligence, Xiamen University, China 2Department of Information Engineering and Computer Science, University of Trento, Italy 3School of Computer Science, Minnan Normal University, China |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described in this paper. |
| Open Datasets | Yes | Datasets. Experiments are conducted on two VT-Re ID datasets. SYSU-MM01 (Wu et al. 2017) are captured by four RGB cameras and two thermal cameras, respectively. Reg DB (Nguyen et al. 2017) comprises 4,120 RGB images and 4,120 infrared images of 412 identities, collected from one RGB camera and one infrared camera. |
| Dataset Splits | Yes | The training set contains 22,258 RGB images and 11,909 infrared images of 395 identities. The testing set involves 3,803 query (infrared) images and 301 gallery (RGB) images of 96 identities. For evaluation, we equally divide Reg DB into the training and testing sets. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, CPLEX 12.4) needed to replicate the experiment. It only mentions general components without versioning. |
| Experiment Setup | Yes | The input images are resized to 384 192 3. The SGD optimizer is used for training and the initial learning rate is set to 0.1 with a warm-up strategy (Luo et al. 2019). We divide the learning rate by 10 after 20 and 50 epochs. We train the model for a total of 80 epochs. For the hyper-parameters of the proposed model, the features are divided into 6 parts (i.e. K = 6). we set α (Eq. 9) to 0.2 and 1.0 for SYSU-MM01 and Reg DB, respectively. γ1 5 (Eq. 13) are to {1, 1, 0.1, 2, 0.1} for SYSU-MM01 and {3, 2, 0.4, 1, 0.6} for Reg DB, respectively. During testing, we set β (Eq. 14) to 0.7 and 0.5 for SYSU-MM01 and Reg DB, respectively. |