PLR-OSNet for Person Re-identification

Learning Diverse Features with Part-Level Resolution for Person Re-Identication, PRCV2020.

Introduction

• The code is developed based on Pytorch framework.
• Our proposed method has achieved the state-of-the-art results on the three popular person Re-ID datasets.[link]

Results(without re-rank)

	rank-1	mAP
Market1501	95.6	88.9
DukeMTMC	91.6	81.2
CUHK03-Labeled	84.6	80.5
CUHK03-Detected	80.4	77.2

Get Stared

1. clone code to your own folder

   git clone https://github.com/AI-NERC-NUPT/PLR-OSNet.git
   

2. Install prerequisites

   • pytorch >=1.1.0
   • torchvision >= 0.3.0
   • yacs >= 0.1.6
   • tb-nightly>=2.0.0
   • Cython >= 0.29.12
   • pytorch-ignite>=0.1.2

3. Prepare Datasets

   You can create a directory to store reid datasets under this repo via

   cd PLR-OSNet
   mkdir data

   (1) Market1501

   The data structure would like:

   data
       market1501
           Market-1501-v15.09.15
               bounding_box_train/
               bounding_box_test/
               query/

   (2) DukeMTMC-ReID

   The data structure would like:

   data
       dukemtmc-reid
           DukeMTMC-reID
               bounding_box_train/
               bounding_box_test/
               query/

   (3) CUHK03

   The data structure would like:

   data 
     cuhk03
         CUHK03_labeled
             bounding_box_train/
             bounding_box_test/
             query/
         CUHK03_detected
             bounding_box_train/
             bounding_box_test/
             query/

Train

scripts/train.sh

python ../main.py \
	--config-file ../configs/im_plr_osnet_triplet_cuhk03_256x128.yaml \
	--transforms random_flip random_erase\
	--root ./data/ \
	--gpu-devices 4

The above train.sh is used for training cuhk03 dataset. You can modify the config file to train other datasets.

market1501 --> configs/im_plr_osnet_triplet_market1501_256x128 .yaml
dukemtmcreid --> configs/im_plr_osnet_triplet_dukemtmcreid_256x128 .yaml

Finally ,you can run train.sh file for training.

cd scripts/train.sh
bash train.sh

Datasets

• Market1501

  https://www.cvfoundation.org/openaccess/content_iccv_2015/papers/Zheng_Scalable_Person_Re-Identification_ICCV_2015_paper.pdf

• DukeMTMC-reID

  https://arxiv.org/abs/1701.07717

• CUHK03

  https://www.cvfoundation.org/openaccess/content_cvpr_2014/papers/Li_DeepReID_Deep_Filter_2014_CVPR_paper.pdf