lhf12278/SIMFGA

Official Code: Spatio-temporal Information Mining and Fusion Feature-Guided Modal Alignment for Video-based Visible-Infrared Person Re-identification

Spatio-temporal Information Mining and Fusion Feature-Guided Modal Alignment for Video-based Visible-Infrared Person Re-identification

Pipeline

Requirements

Installation

We use /torch >=1.12.0 / 24G RTX3090 for training and evaluation.

pip install -r SIMFGL/requirements.txt

Prepare Datasets

1. Download BUPTCampus from baidu disk. The file structure should be:

path_to_dataset
|—— DATA
|—— data_paths.json
|—— gallery.txt
|—— query.txt
|—— train.txt
|—— train_auxiliary.txt

2. Download VCM-HITSZ from baidu disk. The file structure should be:

path_to_dataset
|—— ID
|—— info

Training and Evaluation

1.Train and test on the VCM-HITSZ dataset by first navigating to the VCM-HITSZ directory and running train.py or test.py.

python train.py/test.py

1.Train and test on the BUPTCampus dataset by first navigating to the BUPTCampus directory and running train.py or test.py.

python train.py/test.py

For direct testing, please download our prepared checkpoints and extracted features from baidu disk.

Acknowledgement

A large part of codes are borrowed from MITML. Thanks for their excellent work!

Contact

If you have any questions, please feel free to contact me. (zuozhigang2024@163.com).