Official pytorch implementation of Context-Aware Unsupervised Clustering for Person Search at BMVC 2021.
-
Clone the repository
git clone https://github.com/VIP-Lab-UNIST/CUCPS.git && cd CUCPS
-
Build environment with conda
conda env create --prefix <your_conda_env_path> -f environment.yml
-
Set the Datasets
Download CUHK-SYSU and PRW to your location and set the dataset path in
./lib/datasets/__init__.py
.
Set the save path(--path) in the .sh files.
-
Train CUHK-SYSU
./train-cuhk.sh
-
Train PRW
./train-prw.sh
Add the checkpoint that you want to evaluate on the --checkpoint_name
option in test-*.sh files. The model shold be with args.json
file.
-
Test CUHK-SYSU
./test-cuhk.sh
-
Test PRW(regular gallery)
./test-prw.sh
-
Test PRW(multi-view gallery)
Set
ignore_cam_id=False
andremove_unlabel=False
ofsearch_performance_calc
function onruns/test.py
./test-prw.sh
The result will be saved in same directory with checkpoint file.
Datasets | mAP(%) | Rank-1(%) |
---|---|---|
CUHK-SYSU | 81.1 | 83.2 |
PRW(regular gallery) | 41.7 | 86.0 |
PRW(multi-view gallery) | 36.6 | 64.9 |
@article{cucps,
title={Context-Aware Unsupervised Clustering for Person Search},
author={Byeong-Ju Han*, Kuhyeun Ko* and Jae-Young Sim},
booktitle = {British Machine Vision Conference (BMVC)},
year={2021},
}