About

This is the official implementation of paper Proxy Task Learning For Cross-Domain Person Re-Identification, ICME 2020.

@inproceedings{huang2020proxy,
  title={Proxy Task Learning For Cross-Domain Person Re-Identification},
  author={Huang, Houjing and Chen, Xiaotang and Huang, Kaiqi},
  booktitle={2020 IEEE International Conference on Multimedia and Expo (ICME)},
  pages={1--6},
  year={2020},
  organization={IEEE}
}

Requirements

• Python 2.7
• Pytorch 1.0.0
• Torchvision 0.2.1
• No special requirement for sklearn version

Dataset Path

Prepare datasets to have following structure:

${project_dir}/dataset
    market1501
        Market-1501-v15.09.15
        Market-1501-v15.09.15_ps_label
    cuhk03_np_detected_jpg
        cuhk03-np  # Extracted from cuhk03-np.zip, https://pan.baidu.com/s/1RNvebTccjmmj1ig-LVjw7A
        cuhk03-np-jpg_ps_label
    duke
        DukeMTMC-reID
        DukeMTMC-reID_ps_label
    peta
        PETA dataset
            3DPeS
            CAVIAR4REID
            ...
    rap
        RAP_dataset
        RAP_annotation
    pa100k       # Not used in this project
        PA-100K  # https://drive.google.com/drive/folders/0B5_Ra3JsEOyOUlhKM0VPZ1ZWR2M
            annotation.mat
            release_data
                release_data

• Market-1501-v15.09.15_ps_label, cuhk03-np-jpg_ps_label and DukeMTMC-reID_ps_label can be downloaded from Baidu Cloud or Google Drive.
• PETA can be downloaded from http://mmlab.ie.cuhk.edu.hk/projects/PETA.html. By unzip PETA.zip, you will obtain PETA dataset.
• RAP (v2) can be downloaded according to license agreement. Please refer to this repository.

Testing

Trained Baseline and PTL models can be downloaded from Baidu Cloud (password 1bvf) or Google Drive. Place the exp folder under the project directory.

For example,

• Test baseline, Market1501 to CUHK03 or Duke.

  gpu=0 only_test=True dataset=market1501 bash script/train_baseline.sh

  This should give result

  M -> D      [mAP:  29.1%], [cmc1:  49.9%], [cmc5:  63.8%], [cmc10:  69.9%]
  

• Test PTL, Market1501 to Duke.

  gpu=0 only_test=True src_dset=market1501 tgt_dset=duke bash script/train_ptl.sh

  This should give result

  M -> D      [mAP:  36.2%], [cmc1:  57.4%], [cmc5:  71.0%], [cmc10:  75.8%]
  

Prepare Attribute

Train attribute recognition on PETA,

gpu=0 bash script/train_attr.sh

The script will also evaluate on validation set.

Predict soft attribute labels on ReID images,

gpu=0 predict_on_dataset=market1501 bash script/predict_attr.sh;
gpu=0 predict_on_dataset=cuhk03_np_detected_jpg bash script/predict_attr.sh;
gpu=0 predict_on_dataset=duke bash script/predict_attr.sh;

The results will be saved inside ${project_dir}/dataset/predicted_attr/....

The predicted attribute labels can also be downloaded from Baidu Cloud (password 1bvf) or Google Drive.

Train Baseline

Train baseline on Market1501

gpu=0 dataset=market1501 bash script/train_baseline.sh

Proxy Task Learning

For Market1501 -> Duke,

gpu=0 src_dset=market1501 tgt_dset=duke bash script/train_ptl.sh

Ablation Study

Ablation can be found in script/train_ptl.sh.

Visualize Human Parsing (Part Segmentation) Labels

python script/save_colorful_ps_label.py