Home

Awesome

Python 3.7.5 PyTorch 1.3.1 Cuda 9.2

😄 The code of journal version of this work is released IIDS. We strongly recommend you to use our journal version which has a much higher performance.

IICS

Pytorch implementation of Paper "Intra-Inter Camera Similarity for Unsupervised Person Re-Identification" (CVPR 2021)

fig1

Installation

1. Clone code

    git clone git@github.com:SY-Xuan/IICS.git
    cd ./IICS

2. Install dependency python packages

    conda create --name IICS --file requirements.txt

3. Prepare dataset

Download Market1501, DukeMTMC-ReID, MSMT17 from website and put the zip file under the directory like

./data
├── dukemtmc
│   └── raw
|       └──DukeMTMC-reID.zip
├── market1501
|   └── raw
│       └── Market-1501-v15.09.15.zip
|── msmt17
|   └── raw
|       └── MSMT17_V2.zip

Usage

1. Download trained model

2. Evaluate Model

Change the checkpoint path in the ./script/test_market.sh

sh ./script/test_market.sh

3. Train Model

You need to download ResNet-50 imagenet pretrained model and change the checkpoint path in the ./script/train_market.sh

sh ./script/train_market.sh

We also provide a better version of our method which can adaptively determine clustering number by setting a similarity threshold

sh ./script/train_market_threshold.sh

Results

DatasetsmAPRank@1Method
Market150172.9%89.5%original
Market150173.9%90.1%threshold
DukeMTMC-ReID64.4%80.0%original
DukeMTMC-ReID66.2%80.8%threshold
MSMT1726.9%56.4%original
MSMT1731.9%62.6%threshold

fig1 fig1

Citations

If you find this code useful for your research, please cite our paper:

@article{xuan2021intrainter,
      title={Intra-Inter Camera Similarity for Unsupervised Person Re-Identification}, 
      author={Shiyu Xuan and Shiliang Zhang},
      year={2021},
      journal={arXiv preprint arXiv:2103.11658},
}

Contact me

If you have any questions about this code or paper, feel free to contact me at shiyu_xuan@stu.pku.edu.cn.

Acknowledgement

Codes are built upon open-reid.