SYSU-MM02 is a new visible-infrared person re-identification dataset constructed from untrimmed videos.
Please send a signed dataset release agreement copy to wuanc@mail.sysu.edu.cn.
If your application is passed, we will send the download link of the dataset.
The pedestrian captured in "examples" folder of SYSU-MM02 has signed a privacy license to allow the images to be used for scientific research and shown in research papers.
The other pedestrians are not allowed to be shown publicly, whose faces are masked to avoid privacy problem.
https://github.com/wuancong/CMAM
If you use the dataset, please cite the following paper:
Ancong Wu, Chengzhi Lin, Wei-Shi Zheng. Asymmetric Mutual Learning for
Unsupervised Transferable Visible-Infrared Re-Identification.
IEEE Transactions on Circuits and Systems for Video Technology (IEEE TCSVT),
2024.
The SYSU-MM02 dataset is an RGB-Infrared (IR) multi-modality pedestrian dataset for unsupervised cross-modality person re-identification. The training set is constructed by pedestrian detection on untrimmed videos, which contains more noises than the training sets of existing manually labeled visible-infrared pedestrian benchmark datasets. The training set contains 7,440 visible images and 13,614 near infrared images. The testing set contains 2,360 visible images and 2,360 near infrared images of 118 identities.
Here are some examples of SYSU-MM02:
If you have any questions, please feel free to contact wuanc@mail.sysu.edu.cn.
Author homepage: isee-ai.cn/~wuancong.