Learning Progressive Modality-Shared Transformers for Effective Visible-Infrared Person Re-identification
Authors: Hu Lu, Xuezhang Zou, Pingping Zhang
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on SYSU-MM01 and Reg DB datasets show that our proposed framework performs better than most state-of-the-art methods. |
| Researcher Affiliation | Academia | Hu Lu1, Xuezhang Zou1, Pingping Zhang2* 1School of Computer Science and Communication Engineering, Jiangsu University 2School of Artificial Intelligence, Dalian University of Technology |
| Pseudocode | No | The paper describes the methods and loss functions using text and mathematical equations but does not provide structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | For model reproduction, we release the source code at https://github.com/hulu88/PMT. |
| Open Datasets | Yes | In this work, we follow previous methods and conduct experiments on two public VI-Re ID datasets. SYSU-MM01 (Wu et al. 2017) has a total of 286,628 visible images and 15,792 infrared images with 491 different person identities. ... Reg DB (Nguyen et al. 2017) contains a total of 412 different person identities. |
| Dataset Splits | Yes | SYSU-MM01... The training set contains 22,258 visible images and 11,909 infrared images of 395 persons, and the test set contains images of another 96 different identities. ... Reg DB... randomly select all images of 206 identities for training and the remaining 206 identities for testing. |
| Hardware Specification | Yes | Our proposed method is implemented with the Huawei-Mindspore toolbox and one NVIDIA RTX3090 GPU. |
| Software Dependencies | No | The paper mentions 'Huawei-Mindspore toolbox' but does not specify its version number or any other software dependencies with their versions. |
| Experiment Setup | Yes | All person images are resized to 256 128 with horizontal flipping and random erasing for data augmentation. For infrared images, color jitter and gaussian blur are additionally applied. The batch size is set to 64, containing a total of 8 different identities. For each identity, 4 visible images and 4 infrared images are sampled. We adopt Adam W optimizer with a cosine annealing learning rate scheduler for training. The basic learning rate is set to 3e 4 and weight decay is set to 1e 4. We train 24 epochs for the SYSU-MM01 and 36 epochs for the Reg DB. For both datasets, the epoch t of the first stage is set to 6, the trade-off parameters λ1 and λ2 are set to 0.5, and the margin parameter m is set to 0.1. |