PP-IPG

Lifelong Visible-Infrared Person Re-Identification with Prompt Pool and Instance-level Prompt Generator

1. Prepare datasets

1. RegDB can be downloaded at this website
2. SYSU-MM01 is released at this link. SYSU-MM01
3. LLCM is released at this link. LLCM
4. HITSZ-VCM is released this link. HITSZ-VCM

You need run python pre_process_sysu.pybefore training.

2. Data sturture:

Please prepare the datasets in below structure:

data_path/
├── RegDB/
│   ├── idx/
│   ├── Thermal/
│   └── Visible/
├── SYSU-MM01/
│   ├── ori_data/
│   ├── train_ir_resized_img.npy
│   ├── train_ir_resized_label.npy
│   ├── train_rgb_resized_img.npy
│   └── train_rgb_resized_label.npy
├── LLCM/
│   ├── idx/
│   ├── nir/
│   ├── test_nir/
│   ├── test_vis/
│   └── vis/
├── HITSZ-VCM/
│   ├── info/
│   ├── Test/
│   └── Train/

3. Training

After preparing dataset in the above structure, you can start training with below command:

python Continual_train.py --data_dir your_data_path

Notice:

1. You may also need modify the data path in Class VCM in data_manager.py.

2. The logs will be saved in ./logs/ and model will be saved in ./save_model/

3. You may need to download the ImageNet pretrained transformer model ViT-Base and ajust the pretrain-path.

4. Testing

When the training is finished, you may use the following command to test the model.

python test.py --resume last_stage_checkpoint.tar

Acknowledgements

Most of the code are based on KRKC and AGW.