Home

Awesome

Weakly Supervised Text-based Person Re-Identification

Note: We have open-sourced the trained model and the code necessary to the inference part, based on which you can easily reproduce the performance reported in the paper under a weakly supervised setting.

Retrieval Results on the CUHK-PEDES dataset

Quantitative_results_1

Requirement

Data Preparation

Testing

  1. Download the trained model from Google Drive.

  2. Conduct the blow command:

sh scripts/run_test_res.sh

The result should be around:

top-1 = 57.10%
top-5 = 78.14%
top-10 = 85.23%

Citation

If you find this work useful in your research, please consider citing:

@InProceedings{Zhao_2021_ICCV,
    author    = {Zhao, Shizhen and Gao, Changxin and Shao, Yuanjie and Zheng, Wei-Shi and Sang, Nong},
    title     = {Weakly Supervised Text-Based Person Re-Identification},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    year      = {2021},
}

Acknowledgement

Our code is largely based on CMPL, and we thank the authors for their implementation. Please also consider citing their wonderful code base.

@inproceedings{ying2018CMPM,
    author = {Ying Zhang and Huchuan Lu},
    title = {Deep Cross-Modal Projection Learning for Image-Text Matching},
    booktitle = {ECCV},
    year = {2018}}