Infrared-Visible Cross-Modal Person Re-Identification with an X Modality
Authors: Diangang Li, Xing Wei, Xiaopeng Hong, Yihong Gong4610-4617
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments are performed on two challenging datasets SYSU-MM01 and Reg DB to evaluate the proposed XIV-Re ID approach. Experimental results show that our method considerably achieves an absolute gain of over 7% in terms of rank 1 and m AP even compared with the latest state-of-the-art methods. |
| Researcher Affiliation | Academia | Diangang Li,1 Xing Wei,1 Xiaopeng Hong,1,3 Yihong Gong2 1Faculty of Electronic and Information Engineering, Xi an Jiaotong University 2School of Software Engineering, Xi an Jiaotong University 3Research Center for Artificial Intelligence, Peng Cheng Laboratory |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or link to open-source code for the described methodology. |
| Open Datasets | Yes | We perform experiments on two publicly available datasets SYSU-MM01 and Reg DB. SYSU-MM01 is a challenging, large-scale dataset dedicated to infrared-visible cross-modal person Re ID, collected by four visible cameras and two near-infrared ones (Wu et al. 2017). Reg DB is constructed by using a pair of aligned visible and infrared cameras (Nguyen et al. 2017). |
| Dataset Splits | Yes | The dataset is divided into a training set with 395 identities and a testing set with 96 identities. The training set consists of 22,258 visible images and 11,909 infrared images, while the query set and the gallery set contain 3,803 infrared images and 301 randomly sampled visible images, respectively. Following (Ye et al. 2018a; Wang et al. 2019c), we randomly split the dataset into two halves, one for training and the other for testing. Training set consists of 2,060 visible images and 2,060 infrared images. |
| Hardware Specification | Yes | The batch size M for each modality is set to be 48 on one single TITAN Xp GPU, resulting a total mini-batch of 144. |
| Software Dependencies | No | The proposed method is implemented with Py Torch. No specific version number for PyTorch or other software dependencies is provided. |
| Experiment Setup | Yes | We adopt the Adam optimizing method and the initial learning rate is set to be 0.00035 with warm-up strategy. The weight decay is set to be 0.0005. The batch size M for each modality is set to be 48 on one single TITAN Xp GPU, resulting a total mini-batch of 144. We set the training epoch to 120. And the learning rate decays at 40th and 70th epoch with a decay factor of 0.1. The margin parameter α1 in LC is set to be 0.5 while the margin parameter α2 in LM is set to be 0.3. The trade-off hyperparameter λ between two modality constraints in Eq. 11 is set to be 0.1. |