This repository contains the code for our paper Unlabeled Samples Generated by GAN Improve the Person Re-identification Baseline in vitro.

The first stage is to generate fake images by DCGAN. We used the code provided in https://github.com/carpedm20/DCGAN-tensorflow and modify some hyper-parameters at https://github.com/layumi/DCGAN-tensorflow. You can directly use my forked code.

For more reference, you can find our modified training code and generating code in ./DCGAN. We wrote a detailed README. If you still has some question, feel free to contact me ([email protected]).

The second stage is to combine the original data and generated data to train the network. This repos includes the baseline code and the three different methods in the paper.

• You can find more detailed code for proposed loss in [forward code] [backward code]. (We write the label smooth loss first and then extend it to LSRO. Here we also provide a brief illustration.)

• Pseudo label is realized in https://github.com/layumi/Person-reID_GAN/blob/master/matlab/%2Bdagnn/Pseudo_Loss.m

(Note that I have included my Matconvnet in this repo, so you do not need to download it again. I has changed some codes comparing with the original version. For example, one of the difference is in /matlab/+dagnn/@DagNN/initParams.m. If one layer has params, I will not initialize it again, especially for pretrained model.)

You just need to uncomment and modify some lines in gpu_compile.m and run it in Matlab. Try it~

(The code does not support cudnn 6.0. You may just turn off the Enablecudnn or try cudnn5.1)

If you fail in compilation, you may refer to http://www.vlfeat.org/matconvnet/install/

Download Market1501 Dataset We take Market1501 as an example in this repos and you can easily extend it to other datasets.

1. Make a dir called data by typing mkdir ./data.

2. Download ResNet-50 model pretrained on Imagenet. Put it in the data dir.

1. Add your dataset path into prepare_data.m and run it. Make sure the code outputs the right image path.

2. Run train_id_net_res_market_new.m.

1. Add your generated data path into prepare_data_gan.m and run it. It will add generated image path into the original image database.

2. Run train_id_net_res_market_K_1.m for training extra-class method.

Or run train_id_net_res_market_lsro.m for training the proposed method.

Or run train_id_net_res_market_pseudo.m for training the pseudo-label method.

(What's new: I also include train_id_net_res_2stream_gan.m for training the code with the method proposed in my another paper. I do not import all files, and you may find the missing code in https://github.com/layumi/2016_person_re-ID. )

1. Run test/test_gallery_query_crazy.m to extract the features of images in the gallery and query set. They will store in a .mat file. Then you can use it to do evaluation.
2. Evaluate feature on the Market-1501. Run evaluation/zzd_evaluation_res_faster.m.

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

@inproceedings{zheng2017unlabeled,
  title={Unlabeled Samples Generated by GAN Improve the Person Re-identification Baseline in vitro},
  author={Zheng, Zhedong and Zheng, Liang and Yang, Yi},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  year={2017}
}

1. 2stream Person re-ID
2. Pedestrian Alignment Network