ReID-Label-Noise

Demo Code for refining person re-identification model under label noise used in [1] and [2].

Highlight

The goal of this work is to learn a robust Re-ID model against different noise types. We introduce an online co-refining (CORE) framework with dynamic mutual learning, where networks and label predictions are online optimized collaboratively by distilling the knowledge from other peer networks.

1. Prepare the Dataset

Note that the demo code use a re-oganized file structure so that the code can be seamlessly applied on three datasets, including Market1501, Duke-MTMC and CUHK03 datasets. The detailed description can be found in this website.

training/
   |--id 1/
      |--img 001001/
      |--img 001002/
   |--id 2/  
      |--img 002001/
      |--img 002002/

query/

gallery/
...

2. Running the Code.

Train a model by

python train_core.py --dataset market --batchsize 32 --noise_ratio 0.2 --lr 0.01 --pattern

--dataset: which dataset "market" , "duke" or "cuhk03".
--batchsize: batch training size.
--noise_ratio: 0.2
--lr: initial learning rate.
--pattern: "patterned noise" or "random noise".
--gpu: which gpu to run.

You need mannully define the data path first.

Parameters: More parameters can be found in the script.

Training Model: The training models will be saved in `checkpoint/".

3. Citation

Please kindly cite this paper in your publications if it helps your research:

@article{tip21core,
  title={Collaborative Refining for Person Re-Identification with Label Noise},
  author={Ye, Mang and Li, He and Du, Bo and Shen, Jianbing and Shao, Ling and Hoi, Steven C. H.},
  journal={IEEE Transactions on Image Processing (TIP)},
  year={2021},
}

@article{tifs20noisy,
  title={PurifyNet: A Robust Person Re-identification Model with Noisy Labels},
  author={Ye, Mang and Yuen, Pong C.},
  journal={IEEE Transactions on Information Forensics and Security (TIFS)},
  volume={15},
  pages={2655--2666},
  year={2020},
}

4. References.

[1] M. Ye, H. Li, B. Du, J. Shen, L. Shao, and S. C., Hoi. Collaborative Refining for Person Re-Identification with Label Noise. IEEE Transactions on Image Processing (TIP), 2021.