A Simple Codebase for Clothes-Changing Person Re-identification (Single-GPU Version)

Based on: Clothes-Changing Person Re-identification with RGB Modality Only (CVPR, 2022)

---

✅ Modified Version Highlights:

• Fully compatible with single-GPU setups
• Removed all torch.distributed dependencies
• Removed apex (AMP) requirement
• Refactored main.py to main_single_gpu.py for clarity
• Simplified dataloaders and samplers for easier training/testing

---

🛠 Requirements

• Python 3.10+
• PyTorch
• yacs
• h5py
• scipy
• torchvision

✅ You no longer need APEX or torch.distributed to run this repo.

---

📦 Dataset: CCVID (optional)

• [BaiduYun] password: q0q2
• [GoogleDrive]

You can also use LTCC, PRCC, VC-Clothes, or DeepChange datasets (configurable).

---

🚀 Getting Started

1. Clone this repo and install dependencies:

   git clone https://github.com/your-username/Simple-CCReID.git
   cd Simple-CCReID
   pip install -r requirements.txt

2. Modify paths in the config: Edit configs/default_img.py or configs/default_vid.py

   _C.DATA.ROOT = "/path/to/your/data"
   _C.OUTPUT = "/path/to/save/outputs"

3. Run training (single GPU):

   python main_single_gpu.py --cfg configs/res50_cels_cal.yaml --dataset prcc --gpu 0

---

🧪 Evaluation

Model will automatically evaluate every few epochs and print Rank-1 accuracy and other metrics. To run evaluation only:

python main_single_gpu.py --cfg configs/res50_cels_cal.yaml --dataset prcc --gpu 0 --eval

---

📌 Citation

If you use our code or baseline, please cite:

@inproceedings{gu2022CAL,
    title={Clothes-Changing Person Re-identification with RGB Modality Only},
    author={Gu, Xinqian and Chang, Hong and Ma, Bingpeng and Bai, Shutao and Shan, Shiguang and Chen, Xilin},
    booktitle={CVPR},
    year={2022},
}

---

🔗 Related Repositories

• Simple-ReID
• fast-reid
• deep-person-reid
• Pytorch ReID

---

📌 Maintainer: [Anh Khoa Nguyen] Feel free to fork, modify, or contribute back!