3DHPose is an improved version of mvpose.
Please follow the installation and usage steps from the original repo
This project aims to estimate multi-person 3D Human poses from multiple views. Because the original repo achieves high accuracy, the main purpose of this project is to speed up the process as much as possible.
All the backbone networks were changed in order to achieve a better inference time.
Object detector used: CenterNet
ReId module used: OsNet
2D Human pose detector: CPN using only GlobalNet with ResNet50 as the backbone network.
| Method | Time | Accuracy |
|---|---|---|
| Original Repo | 12:21(min) | 96.36% |
| 3DHPose | 02:14(min) | 96.21% |
Both configurations were tested using a Tesla P100 GPU, on Google Colab. The test dataset is the campus one from the original repo.
@article{dong2019fast,
title={Fast and Robust Multi-Person 3D Pose Estimation from Multiple Views},
author={Dong, Junting and Jiang, Wen and Huang, Qixing and Bao, Hujun and Zhou, Xiaowei},
journal={CVPR},
year={2019}
}