Introduction

Multi Person PoseEstimation By PyTorch

Results

Require

1. Pytorch

2. pytorch-lightning

Installation

1. git submodule init && git submodule update

2. cd pytorch-lightning; python setup.py install

Demo

• Download converted pytorch model.

• Compile the C++ postprocessing: cd lib/pafprocess; sh make.sh

• python demo/picture_demo.py to run the picture demo.

• python demo/web_demo.py to run the web demo.

Evalute

• python evaluate/evaluation.py to evaluate the model on coco val2017 dataset.

• It should have mAP 0.653 for the rtpose, previous rtpose have mAP 0.577 because we do left and right flip for heatmap and PAF for the evaluation. c

Main Results

Download link: rtpose

Development environment

The code is developed using python 3.6 on Ubuntu 18.04. NVIDIA GPUs are needed. The code is developed and tested using 4 1080ti GPU cards. Other platforms or GPU cards are not fully tested.

Quick start

1. Preparation

1.1 Prepare the dataset

• cd training; bash getData.sh to obtain the COCO 2017 images in /data/root/coco/images/, keypoints annotations in /data/root/coco/annotations/, make them look like this:

${DATA_ROOT}
|-- coco
    |-- annotations
        |-- person_keypoints_train2017.json
        |-- person_keypoints_val2017.json
    |-- images
        |-- train2017
            |-- 000000000009.jpg
            |-- 000000000025.jpg
            |-- 000000000030.jpg
            |-- ... 
        |-- val2017
            |-- 000000000139.jpg
            |-- 000000000285.jpg
            |-- 000000000632.jpg
            |-- ...

2. How to train the model

• Modify the data directory in train/train_VGG19.py and python train/train_VGG19.py

Related repository

• CVPR'17, Realtime Multi-Person Pose Estimation.

Network Architecture

• testing architecture 

• training architecture 

Contributions

All contributions are welcomed. If you encounter any issue (including examples of images where it fails) feel free to open an issue.

Citation

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

@INPROCEEDINGS{8486591, 
author={Haoqian Wang and Wang Peng An and X. Wang and L. Fang and J. Yuan}, 
booktitle={2018 IEEE International Conference on Multimedia and Expo (ICME)}, 
title={Magnify-Net for Multi-Person 2D Pose Estimation}, 
year={2018}, 
volume={}, 
number={}, 
pages={1-6}, 
month={July},}

@InProceedings{cao2017realtime,
  title = {Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields},
  author = {Zhe Cao and Tomas Simon and Shih-En Wei and Yaser Sheikh},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2017}
  }