Hierarchical Online Instance Matching for Person Search

This repository hosts our code for our paper Hierarchical Online Instance Matching for Person Seach.

Preparation

1. Clone this repo

   git clone https://github.com/DeanChan/HOIM-PyTorch.git && cd HOIM-PyTorch
   

2. Build environment with conda

   conda create --name HOIM --file requirements.txt
   

Experiments

1. Download CUHK-SYSU and PRW to data/

2. Download the trained model

   mkdir logs && wget https://github.com/DeanChan/HOIM-PyTorch/releases/download/v0.0.0/logs.tar.gz -P logs/ 
   cd logs && tar xvzf logs.tar.gz
   

3. Test

   CUDA_VISIBLE_DEVICES=0 python scripts/test_hoim.py -p logs/cuhk_sysu/
   

4. Train

   CUDA_VISIBLE_DEVICES=0 python scripts/train_hoim.py --debug --lr_warm_up -p ./logs/<your_logging_path>/ --batch_size 5 --nw 5 --w_RCNN_loss_bbox 10.0 --epochs 22 --lr 0.003
   

Citation

@inproceedings{chen2020hoim,
      title={Hierarchical Online Instance Matching for Person Search},
      author={Chen, Di and Zhang, Shanshan and Ouyang, Wanli and Yang, Jian and Schiele, Bernt},
      booktitle={AAAI},
      year={2020}
    }