This is a traditional domain adaptative pedestrian recognition (re-identification) task. The source domain and target domain are fixed and contain entirely different classes. The objective is to achieve possibly high accuracy on the target domain. Submitted ranking lists are evaluated by our online server.

Task Setting

  • This is a domain adaptation task for pedestrian recognition (re-identification).
  • Source domain (part of the training set) is the synthetic Person X dataset (engine). Identity labels are provided
  • Target domain data will be available after finishing the register and confirmation.
  • The target domain consists of three subsets. The first subset forms the other part of the training set, but does not have labels; it is used for style alignment. The second subset is the validation set, where identity labels are provided. The third subset is the test set, where the labels will not be released.

Leader Board

TeamName Rank-1 Rank-5 Rank-10 mAP Rank
SPGAN (PCB ) 0.28971 0.45277 0.538259 0.100823 1
Direct Transfer (PCB) 0.210554 0.347757 0.427441 0.078306 2
Direct Transfer (IDE) 0.170976 0.308179 0.38628 0.065966 3


If you work on this task in your research, please consider citing the following papers.

  author = {Zheng, Liang and Shen, Liyue and Tian, Lu and and Wang, Shengjin and Wang, Jingdong and Tian, Qi},
  title = {Scalable person re-identification: A benchmark},
  booktitle = {Proceedings of the IEEE International Conference on Computer Vision (ICCV)},
  year = {2015}
  author = {Sun, Xiaoxiao and Zheng, Liang},
  title = {Dissecting Person Re-identification from the Viewpoint of Viewpoint},
  booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2019}
  author = {Deng, Weijian and Zheng, Liang and Ye, Qixiang and Kang, Guoliang and Yang, Yi and Jiao, Jianbin},
  title = {Image-image domain adaptation with preserved self-similarity and domain-dissimilarity for person re-identification},
  booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2018}