MAE

Our paper Multi-Attribute Enhancement Network for Person Search has been accepted by IJCNN2021. Our code is stored in this repository.

Preparation

1. Clone this repo:

 https://github.com/chenlq123/MAE.git && cd MAE

2. Requirements

   Python 3.6 with all requirements.txt (except for the first line) dependencies installed. To install run:

 pip install -r requirements.txt

3. Download the Trained Model from the Release.

4. Download the dataset of CUHK-SYSU and PRW to the data/.

5. Download the Attribute Label from the Release. And put them in dataset, respectively.

6. Before Training

   You need to modify the path in the lib/datasets/init.py and lib/model/faster_rcnn_mae.py.

7. Test for CUHK-SYSU or PRW

     CUDA_VISIBLE_DEVICES=0 python scripts/test_MAE.py -p pre_logs/cuhk_sysu/

     CUDA_VISIBLE_DEVICES=0 python scripts/test_MAE.py -p pre_logs/prw/  --dataset PRW

8. Train for CUHK-SYSU or PRW

   CUDA_VISIBLE_DEVICES=0 python scripts/train_MAE.py --debug --lr_warm_up -p ./logs/ --batch_size 2 --nw 2 --w_RCNN_loss_bbox 10.0 --epochs 22 --lr 0.003 --lr_decay_step 8

   CUDA_VISIBLE_DEVICES=0 python scripts/train_MAE.py --debug --lr_warm_up  --dataset PRW -p ./logs/ --batch_size 2 --nw 2 --w_RCNN_loss_bbox 10.0 --epochs 16 --lr 0.003 --lr_decay_step 8

Result

Citation

@inproceedings{chen2021mae,
  title={Multi-Attribute Enhancement Network for Person Search},
  author={Lequan Chen, Wei Xie, Zhigang Tu, Jinglei Guo, Yaping Tao, Xinming Wang},
  booktitle={IJCNN},
  year={2021}
}