FedReID

Code for ACMMM 2020 oral paper - Performance Optimization for Federated Person Re-identification via Benchmark Analysis

Personal re-identification is an important computer vision task, but its development is constrained by the increasing privacy concerns. Federated learning is a privacy-preserving machine learning technique that learns a shared model across decentralized clients. In this work, we implement federated learning to person re-identification (FedReID) and optimize its performance affected by statistical heterogeneity in the real-world scenario.

Algorithm: Federated Partial Averaging (FedPav)

Prerequisite

• Install the libraries listed in requirements.txt

  pip install -r requirements.txt
  

Datasets preparation

We use 9 popular ReID datasets for the benchmark.

You can obtain the datasets from awesome-reid-dataset

Dataset folder structure after preprocessing is provided here

You can follow the following steps to preprocess datasets:

1. Download all datasets to data_preprocess/data folder.
2. We provide the Json files for spliting the small datasets. (We haven't officially release the split.json files. Please send an email with short introduction to request for them.)
3. Run the following script to prepare all datasets:

   python prepare_all_datasets.py
   

4. Move the data folder to the root directory.

   move data_preprocess/data ./
   

5. For federated-by-identity scenario:

   python split_id_data.py
   

6. For federated-by-camera scenario:

   python split_camera_data.py
   

7. For merging all datasets to do merge training, you can use rename_dataset.py and mix_datasets.py.

Run the experiments

Remember to save the log file for later use!

• Run Federated Partial Averaging (FedPav):

  python main.py
  

• Run FedPav with knowledge distillation (KD):

  python main.py --kd --regularization
  

• Run FedPav with cosine distance weight (CDW):

  python main.py --cdw
  

• Run FedPav with knowledge distillation and cosine distance weight:

  python main.py --cdw --kd --regularization
  

Citation

@inproceedings{zhuang2020performance,
  title={Performance Optimization of Federated Person Re-identification via Benchmark Analysis},
  author={Zhuang, Weiming and Wen, Yonggang and Zhang, Xuesen and Gan, Xin and Yin, Daiying and Zhou, Dongzhan and Zhang, Shuai and Yi, Shuai},
  booktitle={Proceedings of the 28th ACM International Conference on Multimedia},
  pages={955--963},
  year={2020}
}

Maintainers

• Weiming Zhuang, Nanyang Technological University.
• Xin Gan, Nanyang Technological University.
• Daiying Yin, Nanyang Technological University. (Contributor)