vikiQiu / PCB_RPP_for_reID

Part-based Convolutional Baseline for Person Retrieval and the Refined Part Pooling

Code for the paper Beyond Part Models: Person Retrieval with Refined Part Pooling (and A Strong Convolutional Baseline).

This code is ONLY released for academic use.

Preparation

Prerequisite: Python 2.7 and Pytorch 0.3+

1. Install Pytorch

2. Download dataset a. Market-1501 BaiduYun b. DukeMTMC-reIDBaiduYun (password:bhbh) c. Move them to ~/datasets/Market-1501/(DukeMTMC-reID)

train PCB

sh train_PCB.sh With Pytorch 0.4.0, we shall get about 93.0% rank-1 accuracy and 78.0% mAP on Market-1501.

train RPP

sh train_RPP.sh With Pytorch 0.4.0, we shall get about 93.5% rank-1 accuracy and 81.5% mAP on Market-1501.

Citiaion

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

@inproceedings{sun2018PCB,
  author    = {Yifan Sun and
               Liang Zheng and
               Yi Yang and
			   Qi Tian and
               Shengjin Wang},
  title     = {Beyond Part Models: Person Retrieval with Refined Part Pooling (and A Strong Convolutional Baseline)},
  booktitle   = {ECCV},
  year      = {2018},
}