VI-Diff: Unpaired Visible-Infrared Translation Diffusion Model for Single Modality Labeled Visible-Infrared Person Re-identification

This is experiment code. Will clean up later.

datasets: put visible/infrared images to trainA/trainB

for sysu-mm01: rename images in format like 'cam1_0001_0002.jpg'

training/testing:

use bash scripts to train/test.

regdb: extract Hed edge, and modify the edge image path in 'guided_diffusion/image_datasets.py';
       edge image name same as original image.

modify 'process_gpu_dict' in 'guided_diffusion/dist_util.py' to use your GPU cluster.

Citations

@misc{huang2023vidiff,
    title={VI-Diff: Unpaired Visible-Infrared Translation Diffusion Model for Single Modality Labeled Visible-Infrared Person Re-identification}, 
    author={Han Huang and Yan Huang and Liang Wang},
    year={2023},
    eprint={2310.04122},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}