Masked Attribute Description Embedding for Cloth-Changing Person Re-identification.

The official repository for Masked Attribute Description Embedding for Cloth-Changing Person Re-identification.

Pipeline

Ablation Study of MADE

Requirements

• Python 3.7
• Pytorch 1.13.1
• yacs
• timm
• spacy

Prepare Datasets

Download the cloth-changing person Re-ID datasets, PRCC, LTCC, Celeb-reID-light and LaST. Then add attributes file(Get from SOLIDER) in each dataset file. Or download attributes files directly(remember to modify the path in file), PRCC, LTCC, Celeb-reID-light, LaST.

Data
├── PRCC
│   └── rgb ..
│   └── sketch ..
│   └── PAR_PETA_105.txt
├── LTCC
│   └── train ..
│   └── query ..
│   └── test ..
│   └── PAR_PETA_105.txt 
├── Celeb-reID-light
│   └── train ..
│   └── query ..
│   └── gallery ..
│   └── PAR_PETA_105.txt 
├── LaST
│   └── train ..
│   └── val ..
│   └── test ..
│   └── PAR_PETA_105.txt 

Training

We utilize 2 GPUs for training. Replace _C.DATA.ROOT in config/defaults.pywith your own data path.

CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 --master_port 6666 train.py --config_file ${1} DATA.ADD_META ${2} DATA.MASK_META ${3} MODEL.DIST_TRAIN True

Arguments

• ${1}: config file path.
• ${2}: whether adding attributes.
• ${3}: whether masking clothing-relevant attribute.

or you can directly train with following yml and commands:

# prcc
CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 --master_port 6666 train.py --config_file configs/prcc/eva02_l_maskmeta_random.yml MODEL.DIST_TRAIN True
# ltcc
CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 --master_port 6666 train.py --config_file configs/ltcc/eva02_l_maskmeta_random.yml MODEL.DIST_TRAIN True
# Celeb_reID_light
CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 --master_port 6666 train.py --config_file configs/Celeb_light/eva02_l_maskmeta_random.yml MODEL.DIST_TRAIN True
# last
CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 --master_port 6666 train.py --config_file configs/last/eva02_l_maskmeta_random.yml MODEL.DIST_TRAIN True

Evaluation

CUDA_VISIBLE_DEVICES=1,0 python -m torch.distributed.launch --nproc_per_node=2 --master_port 6666 test.py --config_file 'choose which config to test' TEST.WEIGHT "('your path of trained checkpoints')"

example:

# LTCC
CUDA_VISIBLE_DEVICES=1,0 python -m torch.distributed.launch --nproc_per_node=2 --master_port 6666 test.py --config_file configs/ltcc/eva02_l_maskmeta_random.yml TEST.WEIGHT '../logs/ltcc/eva02_l_meta_best.pth'

Acknowledgement

Codebase from TransReid , CAL

Citation

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

@misc{peng2024masked,
      title={Masked Attribute Description Embedding for Cloth-Changing Person Re-identification}, 
      author={Chunlei Peng and Boyu Wang and Decheng Liu and Nannan Wang and Ruimin Hu and Xinbo Gao},
      year={2024},
      eprint={2401.05646},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}