[CVPR2020] Hi-CMD: Hierarchical Cross-Modality Disentanglement for Visible-Infrared Person Re-Identification
Feel free to visit my homepage and awesome person re-id github page
1-minute version (ENG)
5-minute version (KOR)
<Illustration of our Hierarchical Cross-Modality Disentanglement (Hi-CMD) concept>
conda create -n env_name python=3.6
source activate env_name
conda install pytorch torchvision cudatoolkit=10.0 -c pytorch
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
git clone https://github.com/Cadene/pretrained-models.pytorch.git
cd pretrained-models.pytorch
python setup.py install
./HiCMD/data/RegDB_01/
sh train.sh
./HiCMD/pretrained/checkpoints/
sh test.sh
Metric | Value |
---|---|
Rank1 | 70.44% |
Rank5 | 79.37% |
Rank10 | 85.15% |
Rank20 | 91.55% |
mAP | 65.93% |
MATLAB is required for evaluating SYSU-MM01 (official code).
Download SYSU_features
./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.
Metric | Value |
---|---|
Rank1 | 34.94% |
Rank5 | 65.48% |
Rank10 | 77.58% |
Rank20 | 88.38% |
mAP | 35.94% |
If you want to experiment with all sets of RegDB, download the entire dataset:
If you want to experiment with SYSU-MM01, download the official dataset:
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
The code is based on the PyTorch implementation of the Person_reID_baseline_pytorch, Cross-Model-Re-ID-baseline, MUNIT, DGNET, SYSU-evaluation.
@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}
}
[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.