Towards Fewer Labels: Support Pair Active Learning for Person Re-identification (SPAL)

Paper

Requirements

Installation

git clone https://github.com/dapeng/SPAL.git
cd SPAL
python setup.py develop

Prepare Datasets

cd examples && mkdir data

Download the person datasets Market-1501, MSMT17. Then unzip them under the directory like

SPAL/examples/data
├── market1501
│   └── Market-1501-v15.09.15
└── msmt17
    └── MSMT17_V1

Prepare ImageNet Pre-trained Models

• ImageNet-pretrained models for ResNet-50 will be automatically downloaded in the python script.

• When training with the backbone of IBN-ResNet, you need to download the ImageNet-pretrained model from this link and save it under the path of logs/pretrained/.

  mkdir logs && cd logs
  mkdir pretrained
  

  The file tree should be

  SPAL/logs 
  └── pretrained
      └── resnet50_ibn_a.pth.tar
  

Training

We utilize 4 GPUs for training. To train the model(s) in the paper, run this command:

CUDA_VISIBLE_DEVICES=0,1,2,3 python examples/train.py -d $DATASET --logs-dir $PATH_OF_LOGS

Note that

• We set the labeling budget to 2n for all the experiments. The actual labeling cost is reported if the cost is less than the budget.

Evaluation

To evaluate the model, run:

CUDA_VISIBLE_DEVICES=0 python examples/test.py -d $DATASET --resume $PATH

Citation

If you find this code useful for your research, please cite our paper

@article{jin2022towards,
  title={Towards Fewer Labels: Support Pair Active Learning for Person Re-identification},
  author={Jin, Dapeng and Li, Minxian},
  journal={arXiv preprint arXiv:2204.10008},
  year={2022}
}

Acknowledgements

Thanks to Yixiao Ge for opening source of her excellent works SpCL.