Awesome
PyTorch implementation of Learning by Aligning (ICCV 2021)
This is an official PyTorch implementation of the paper "Learning by Aligning: Visible-Infrared Person Re-identification using Cross-Modal Correspondences", ICCV 2021.
For more details, visit our project site or see our paper.
Requirements
- Python 3.8<br>
- PyTorch 1.7.1<br>
- GPU memory >= 11GB<br>
Getting started
First, clone our git repository.<br>
git clone https://github.com/cvlab-yonsei/LbA.git
cd LbA
Docker
We provide a Dockerfile to help reproducing our results easily.<br>
Prepare datasets
- SYSU-MM01: download from this link.<br>
- For SYSU-MM01, you need to preprocess the .jpg files into .npy files by running:<br>
python utils/pre_preprocess_sysu.py --data_dir /path/to/SYSU-MM01<br>
- Modify the dataset directory below accordingly.<br>
- L63 of
train.py<br> - L54 of
test.py<br>
- L63 of
- For SYSU-MM01, you need to preprocess the .jpg files into .npy files by running:<br>
Train
-
run
python train.py --method full<br> -
Important:
- Performances reported during training does <ins>not</ins> reflect exact performances of your model. This is due to 1) evaluation protocols of the datasets and 2) random seed configurations.<br>
- Make sure you seperately run
test.pyto obtain correct results to be reported in your paper.<br>
Test
- run
python test.py --method full<br> - The results should be around: <br>
| dataset | method | mAP | rank-1 |
|---|---|---|---|
| SYSU-MM01 | baseline | 49.54 | 50.43 |
| SYSU-MM01 | full | 54.14 | 55.41 |
Pretrained weights
- Download [SYSU-MM01]<br>
- The results should be: <br>
| dataset | method | mAP | rank-1 |
|---|---|---|---|
| SYSU-MM01 | full | 55.22 | 56.31 |
Bibtex
@inproceedings{park2021learning,
title={Learning by Aligning: Visible-Infrared Person Re-identification using Cross-Modal Correspondences},
author={Park, Hyunjong and Lee, Sanghoon and Lee, Junghyup and Ham, Bumsub},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={12046--12055},
year={2021}
}