Exploiting Robust Unsupervised Video Person Re-identification

Implementation of the proposed uPMnet. For the preprint, please refer to [Arxiv].

Getting Started

Requirements

Here is a brief instruction for installing the experimental environment.

> ~/.bashrc

# install other dependencies
$ pip install scipy matplotlib
“>

# install virtual envs
$ conda create -n uPMnet python=2.7 -y
$ conda activate uPMnet

# install tensorflow 1.4.0 with cuda 9.0
$ pip install --ignore-installed --upgrade https://github.com/mind/wheels/releases/download/tf1.4-gpu-cuda9/tensorflow-1.4.0-cp27-cp27mu-linux_x86_64.whl

# install mkl
$ sudo apt install cmake
$ git clone --branch v0.12 https://github.com/01org/mkl-dnn.git
$ cd mkl-dnn/scripts; ./prepare_mkl.sh && cd ..
$ mkdir -p build && cd build && cmake .. && make -j36
$ sudo make install
$ echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib' >> ~/.bashrc

# install other dependencies
$ pip install scipy matplotlib

Convert benchmarks to tfrecords

# Please modify the path in your way
$ bash datasets/convert_data_to_tfrecords.py

Download pre-trained models

The Mobilenet and Resnet models can be downloaded in this link (code: 1upx) and should be put in the checkpoints directory.

Training and Extracting features

$ bash scripts/train_PRID2011.sh # train_iLIDS_VID.sh or train_DukeMTMC-VideoReID.sh

Testing

Use the Matlab to run the following files, evaluation/CMC_PRID2011.m, evaluation/CMC_iLIDS-VID.m, and evaluation/CMC_DukeMTMC_VideoReID.m.

Results in the Paper

The results of PRID2011, iLIDS-VID, and DukeMTMC-VideoReID are provided.

Model	[email protected]	[email protected]	[email protected]
uPMnet	92.0 link (code: xa7z)	63.1 link (code: le2c)	83.6 link (code: e9ja)

You can download these results and put them in the results directory. Then use Matlab to evaluate them.

Acknowledgement

This repository is built upon the repository DAL.

Citation

If you find this project useful for your research, please kindly cite:

@article{zang2021exploiting,
  title={Exploiting Robust Unsupervised Video Person Re-identification},
  author={Zang, Xianghao and Li, Ge and Gao, Wei and Shu, Xiujun},
  journal={arXiv preprint arXiv:2111.05170},
  year={2021}
}