JSIA-ReID
This is the official implementation for JSIA-ReID. Please refer our paper for more details:
[Paper] [Neural Network 2020] Cross-Modality Paired-Images Generation and Augmentation for RGB-Infrared Person Re-Identification
[Paper, Poster] [AAAI2020] Cross-Modality Paired-Images Generation for RGB-Infrared Person Re-Identification
Bibtex
If you find the code useful, please consider citing our paper:
@InProceedings{wang2020cross,
title={Cross-modality paired-images generation and augmentation for RGB-infrared person re-identification},
author={Guan'an, Wang and Yang, Yang and Zhang, Tianzhu and Cheng, Jian and Hou, Zengguang and Tiwari, Prayag and Pandey, Hari Mohan and others},
journal={Neural Networks},
volume={128},
pages={294--304},
year={2020},
publisher={Elsevier}
}
@InProceedings{wang2020crossmodality,
author = "Guan-An {Wang} and Tianzhu {Zhang} and Yang {Yang} and Jian {Cheng} and Jianlong {Chang} and Xu {Liang} and Zengguang {Hou}",
title = {Cross-Modality Paired-Images Generation for RGB-Infrared Person Re-Identification},
booktitle = {AAAI-20 AAAI Conference on Artificial Intelligence},
year = {2020}
}
Dependencies
- Anaconda (Python 3.7)
- PyTorch 1.1.0
- GPU Memory >= 20G
- Memory >= 20G
Dataset Preparation
- SYSU-MM01 Dataset [link]
Train
# train, please replace sysu-mm01-path with your own path
python main.py --dataset_path sysu-mm01-path --mode train
Test with Pre-trained Model
- pretrained model (Google Drive， Baidu Disk(pwd:656y)), please download all the 4 files into a folder.
- test with the pre-trained model
# test with pretrained model, please replace sysu-mm01-path and pretrained-model-path with your own paths
python main.py --dataset_path sysu-mm01-path --mode test --pretrained_model_path pretrained-model-path --pretrained_model_epoch 649
Experimental Results
-
Settings: We trained our model with 2 GTX1080ti GPUs.
-
Comparison with SOTA
License
This repo is released under the MIT License.
Contacts
If you have any question about the project, please feel free to contact with me.
E-mail: [email protected]