Multiple Expert Brainstorming for Domain Adaptive Person Re-identification

Yunpeng Zhai, Qixiang Ye, Shijian Lu, Mengxi Jia, Rongrong Ji, Yonghong Tian. "Multiple Expert Brainstorming for Domain Adaptive Person Re-identification", ECCV 2020

Paper PDF

Setup

Datasets (Market-1501 and DukeMTMC-reID).

Requirements

• PyTorch 1.3.1

Running the experiments

Step:1 Supervised learning in the source domain

bash pretrain.sh <source dataset> <target dataset> <architecture>

For example, (duke->market):

bash pretrain.sh dukemtmc market1501 densenet
bash pretrain.sh dukemtmc market1501 resnet50
bash pretrain.sh dukemtmc market1501 inceptionv3

Step:2 Unsupervised adaptation in the target domain

bash train.sh <source dataset> <target dataset> <architecture-1> <architecture-2> <architecture-3>

For example, (duke->market)

bash train.sh dukemtmc market1501 densenet resnet50 inceptionv3

Step:3 Evaluate in the target domain

bash test.sh <target dataset> <architecture> <checkpoint_path>

For example, (market1501, densenet)

bash test.sh market1501 densenet logs/xxxx/xxxx-MEB-Net/checkpoint.pt.pth

Experiment results

Acknowledgement

Our code is based on open-reid and MMT.

Citation

If you use this method or this code in your research, please cite as:

@article{zhai2020multiple,
  title={Multiple Expert Brainstorming for Domain Adaptive Person Re-identification},
  author={Zhai, Yunpeng and Ye, Qixiang and Lu, Shijian and Jia, Mengxi and Ji, Rongrong and Tian, Yonghong},
  journal={arXiv preprint arXiv:2007.01546},
  year={2020}
}