Feel free to visit my homepage and awesome person re-id github page

Hi-CMD: Hierarchical Cross-Modality Disentanglement for Visible-Infrared Person Re-Identification [CVPR2020 paper]

Presentation video

1-minute version (ENG)

5-minute version (KOR)

<Illustration of our Hierarchical Cross-Modality Disentanglement (Hi-CMD) concept>

Prerequisites

• Ubuntu 18.04
• Python 3.6
• PyTorch 1.0+ (recent version is recommended)
• NVIDIA GPU (>= 8.5GB)
• CUDA 10.0 (optional)
• CUDNN 7.5 (optional)

Getting Started

Installation

• Configure virtual (anaconda) environment

conda create -n env_name python=3.6
source activate env_name
conda install pytorch torchvision cudatoolkit=10.0 -c pytorch

• Install python libraries

conda install -c conda-forge matplotlib
conda install -c anaconda yaml
conda install -c anaconda pyyaml 
conda install -c anaconda scipy
conda install -c anaconda pandas 
conda install -c anaconda scikit-learn 
conda install -c conda-forge opencv
conda install -c anaconda seaborn
conda install -c conda-forge tqdm

• Install "Pretrained models for Pytorch"
  • To use pretrained models

git clone https://github.com/Cadene/pretrained-models.pytorch.git
cd pretrained-models.pytorch
python setup.py install

• Clone this repo:
  • download

Training and testing

• In the actual experiment, a total of 10 experiment sets are used.
• Here is a simple example of running an experiment on only one set (RegDB-01).
• Download [RegDB_01] (for a simple test)
  • The RegDB_01 dataset should be included in './data/'
  • Ex: ./HiCMD/data/RegDB_01/
• You can download the entire sets of RegDB at this [link]
  • If you will use this dataset, please refer to the papers [1,2] below.

Training

sh train.sh

Testing on pretrained model

1) RegDB_01 [1,2]

• Download RegDB_pretrained
  • The pretrained RegDB_01 model should be included in './pretrained/'
  • Ex: ./HiCMD/pretrained/checkpoints/

sh test.sh

• The code provides the following results.

• Note that the performance of the manuscript (Rank1: 70.93%) is obtained by averaging this experiment for all 10 sets.
• If the code is not working, please refer to './pretrained/test_results/net_70000_RegDB_01_(ms1.0)_f1_test_result.txt'

2) SYSU-MM01 [3]

• MATLAB is required for evaluating SYSU-MM01 (official code).

• Download SYSU_features

  • The pretrained SYSU-MM01 features should be included in './eval_SYSU/'
  • Ex: ./HiCMD/eval_SYSU/

• The code provides the following results.

• Correction: In the paper, I wrote, "The SYSU dataset contains 22,258 visible images and 11,909 near-infrared images of 395 identities for training". This value indicates the total of training and validation sets. But, in fact, I only used a training set including 20,284 visible images and 9,927 infrared images of 296 identities. I apologize for the confusion.

• If the code is not working, please refer to './eval_SYSU/results_test_SYSU.txt'

(Optional) If all the files can not downloaded in the above links, please check the below links.

• HiCMD-master.zip
• RegDB_01
• RegDB_original
• RegDB_pretrained
• SYSU_features

(Optional) Additional experiments

• If you want to experiment with all sets of RegDB, download the entire dataset:

  • The RegDB dataset [1, 2] can be downloaded from this link. (The original name is "Dongguk Body-based Person Recognition Database (DBPerson-Recog-DB1)").
  • If you will use this dataset, please refer to the papers [1,2] below.

• If you want to experiment with SYSU-MM01, download the official dataset:

  • The SYSU-MM01 dataset [3] can be downloaded from this website.
  • The authors' official matlab code is used to evaluate the SYSU dataset.
  • If you will use this dataset, please refer to the paper [3] below.

• Change the 'data_name' from 'RegDB_01' to the name of other datasets.

• Process the downloaded data according to the code by python prepare.py.

• Train and test

Acknowledgement

The code is based on the PyTorch implementation of the Person_reID_baseline_pytorch, Cross-Model-Re-ID-baseline, MUNIT, DGNET, SYSU-evaluation.

Citation

@InProceedings{Choi_2020_CVPR,
author = {Choi, Seokeon and Lee, Sumin and Kim, Youngeun and Kim, Taekyung and Kim, Changick},
title = {Hi-CMD: Hierarchical Cross-Modality Disentanglement for Visible-Infrared Person Re-Identification},
booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}

Reference

• [1] D. T. Nguyen, H. G. Hong, K. W. Kim, and K. R. Park. Person recognition system based on a combination of body images from visible light and thermal cameras. Sensors, 17(3):605, 2017.

• [2] Jin Kyu Kang, Toan Minh Hoang, and Kang Ryoung Park, “Person Re-Identification Between Visible and Thermal Camera Images Based on Deep Residual CNN Using Single Input,” IEEE Access, Vol. 7, pp. 57972-57984, May 2019

• [3] A. Wu, W.-s. Zheng, H.-X. Yu, S. Gong, and J. Lai. Rgb-infrared crossmodality person re-identification. In IEEE International Conference on Computer Vision (ICCV), pages 5380–5389, 2017.