daodaofr / AlignPS

Code for CVPR 2021 paper: Anchor-Free Person Search

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

PWC PWC

Update

We extend AlignPS by augmenting it with an ROI-Align head, which further improves the performance. Details can be found here.

Dataset Model mAP Rank1 Config Link
CUHK-SYSU ROI-AlignPS 95.1% 96.0% cfg model
PRW ROI-AlignPS 51.6% 84.4% cfg model

Introduction

This is the implementation for Anchor-Free Person Search in CVPR2021 and its extended version Efficient Person Search: An Anchor-Free Approach.

A brief introduction in Chinese can be found at https://zhuanlan.zhihu.com/p/359617800

demo image

License

This project is released under the Apache 2.0 license.

Installation

This project is developed upon MMdetection, please refer to install.md to install MMdetection.

We utilized mmcv=1.1.5, pytorch=1.7.0

Dataset

Download CUHK-SYSU and PRW.

We provide coco-style annotation in demo/anno.

For CUHK-SYSU, change the path of your dataset and the annotaion file in the config file L3, L38, L43, L48

For PRW, change the paths in these config files: config1 config2

Experiments

  1. Train
sh run_train.sh
  1. Test CUHK-SYSU

Change the paths in L59 and L72 in test_results.py

sh run_test.sh

For ROI-AlignPS, change the paths in test_results_psd.py

sh run_test_roi.sh
  1. Test PRW

Change the paths in L127 and L128 in test_results_prw.py

sh run_test_prw.sh

For ROI-AlignPS, change the paths in test_results_psd_prw.py

sh run_test_roi_prw.sh

Performance

Dataset Model mAP Rank1 Config Link
CUHK-SYSU AlignPS 93.1% 93.4% cfg model
CUHK-SYSU AlignPS+ 94.0% 94.5% cfg model
PRW AlignPS 45.9% 81.9% cfg model
PRW AlignPS+ 46.1% 82.1% cfg model

Citation

If you use this toolbox or benchmark in your research, please cite this project.

@inproceedings{DBLP:conf/cvpr/YanLQBL00021,
  author    = {Yichao Yan and
               Jinpeng Li and
               Jie Qin and
               Song Bai and
               Shengcai Liao and
               Li Liu and
               Fan Zhu and
               Ling Shao},
  title     = {Anchor-Free Person Search},
  booktitle = {{IEEE} Conference on Computer Vision and Pattern Recognition},
  pages     = {7690--7699},
  year      = {2021}
}

@article{DBLP:journals/corr/abs-2109-00211,
  author    = {Yichao Yan and
               Jinpeng Li and
               Jie Qin and
               Shengcai Liao and
               Xiaokang Yang},
  title     = {Efficient Person Search: An Anchor-Free Approach},
  journal   = {CoRR},
  volume    = {abs/2109.00211},
  year      = {2021}
}

About

Code for CVPR 2021 paper: Anchor-Free Person Search

License:Apache License 2.0


Languages

Language:Python 99.8%Language:Shell 0.2%Language:Dockerfile 0.0%