View Code? Open in Web Editor NEW

Pose-guided Inter- and Intra-part Relational Transformer for Occluded Person Re-Identification ACM MM 2021

License: Apache License 2.0

Python 96.84% Makefile 0.03% Cython 3.13%

pirt's Introduction

Pirt

Pose-guided Inter- and Intra-part Relational Transformer for Occluded Person Re-Identification official implement

Introduction

This repository contains the code for the paper: Pose-guided Inter- and Intra-part Relational Transformer for Occluded Person Re-Identification Zhongxing Ma, Yifan Zhao, Jia Li ACM Conference on Multimedia (ACM MM), 2021

Environments

pytorch 1.6.0
python 3.8
pyyaml | yacs | termcolor | tqdm | faiss-cpu | tabulate | tabulate | matplotlib | tensorboard
sklearn | enopis
GTX 2080 Ti * 2
CUDA 10.1

Getting Started

Working directory: /your/path/to/fast-reid/

Traning

python -u tools/train_net.py --config-file configs/Pirt.yml --num-gpus 2 OUTPUT_DIR logs/your/customed/path

Evaluation

python -u tools/train_net.py --eval-only --config-file configs/eval.yml --num-gpus 2 OUTPUT_DIR logs/your/customed/path

The config file of the model are placed at ./configs/Pirt.yml

Datasets

OccludedDuke or Market-1501 datasets shoule be placed at ./datasets/OccludedDuke

See the ./fastreid/data/datasets folder for detailed configuration

Pretrained Models

The pose-guided and resnet50 models should be placed at ../models_zoo/

resnet50. resnet50-ibn. pose_hrnet_w48_256x192.

Citing

@misc{ma2021poseguided,
      title={Pose-guided Inter- and Intra-part Relational Transformer for Occluded Person Re-Identification}, 
      author={Zhongxing Ma and Yifan Zhao and Jia Li},
      year={2021},
      eprint={2109.03483},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}