Home

Awesome

Code for IDE baseline on Market-1501

============= This code was used for experiments with ID-discriminative Embedding (IDE) for Market-1501 dataset.

Thanks Liboyue, give us suggestions for improvement.

If you find this code useful in your research, please consider citing:

@article{zheng2016person,
title={Person Re-identification: Past, Present and Future},
author={Zheng, Liang and Yang, Yi and Hauptmann, Alexander G},
journal={arXiv preprint arXiv:1610.02984},
year={2016}
}

@inproceedings{zheng2015scalable,
title={Scalable Person Re-identification: A Benchmark},
author={Zheng, Liang and Shen, Liyue and Tian, Lu and Wang, Shengjin and Wang, Jingdong and Tian, Qi},
booktitle={Computer Vision, IEEE International Conference on},
year={2015}
}

Requirements: Caffe

Requirements for Caffe and matcaffe (see: Caffe installation instructions)

Installation

  1. Clone the IDE repository
# Make sure to clone with --recursive
git clone --recursive https://github.com/zhunzhong07/IDE-baseline-Market-1501
  1. Build Caffe and matcaffe

    cd $IDE_ROOT/caffe
    # Now follow the Caffe installation instructions here:
    # http://caffe.berkeleyvision.org/installation.html
    make -j8 && make matcaffe
    
  2. Download pre-computed models and Market-1501 dataset

Please download the pre-trained imagenet models and put it in the "data/imagenet_models" folder.
Please download Market-1501 dataset and unzip it in the "market_evaluation/dataset" folder. 

Training and testing IDE model

  1. Training
cd $IDE_ROOT
 # train IDE on CaffeNet
./experiments/market/train_IDE_CaffeNet.sh  
# train IDE ResNet_50
./experiments/market/train_IDE_ResNet_50.sh
# The IDE models are saved under: "out/market_train"
# If you encounter this problem: bash: ./experiments/market/train_IDE_CaffeNet.sh: Permission denied
# Please execute: chmod 777 -R experiments/
  1. Feature Extraction
cd $IDE_ROOT/market_evaluation
Run Matlab: extract_feature.m
# The IDE features are saved under: "market_evaluation/feat"
  1. Evaluation
  Run Matlab: baseline_evaluation_IDE.m

Results

You can download our pre-trained IDE models and IDE features, and put them in the "out_put/market_train" and "market_evaluation/feat" folder, respectively.

Using the models and features above, you can reproduce the results as follows:

Methods  Rank@1mAP
IDE_CaffeNet + Euclidean59.53%32.85%
IDE_CaffeNet + XQDA      62.00%37.55%
IDE_CaffeNet + KISSME61.02%36.72%
IDE_ResNet_50 + Euclidean75.62%50.68%
IDE_ResNet_50 + XQDA76.01%52.98%
IDE_ResNet_50 + KISSME77.52%53.88%

If you add a dropout = 0.5 layer after pool5, you will get a better performance for ResNet_50:

Methods  Rank@1mAP
IDE_ResNet_50 + dropout(0.5) + Euclidean78.92%55.03%
IDE_ResNet_50 + dropout(0.5) + XQDA77.35%56.01%
IDE_ResNet_50 + dropout(0.5) + KISSME78.80%56.13%

Contact us

If you have any questions about this code, please do not hesitate to contact us.

Zhun Zhong

Liang Zheng

Related Repos

Furthermore, you may check the following codes.

  1. re-ranking
  2. 2stream Network for reID
  3. Person re-ID with GAN
  4. Pedestrian Alignment Network
  5. Random-Erasing