CenterHMR: a bottom-up single-shot method for multi-person 3D mesh recovery from a single image

The method achieves ECCV 2020 3DPW Challenge Runner Up. Please refer to arxiv paper for the details!

Installation

Download models

Option 1:

Directly download the full-packed released package CenterHMR.zip from github, latest version v0.0.

Option 2:

Clone the repo:

git clone https://github.com/Arthur151/CenterHMR --depth 1

Then download the CenterHMR data from Google drive or Baidu Drive with password 6hye.

Unzip the downloaded CenterHMR_data.zip under the root CenterHMR/. The layout would be

CenterHMR
  - demo
  - models
  - src
  - trained_models

Set up environments

Please intall the Pytorch 1.6 from the official webset. We have tested the code on Ubuntu and Centos using Pytorch 1.6 only.

Install packages:

cd CenterHMR/src
sh scripts/setup.sh

Please refer to the bug.md for unpleasant bugs. Feel free to submit the issues for related bugs.

Demo

Currently, the released code is used to re-implement demo results. Only 1-2G GPU memery is needed.

To do this you just need to run

cd CenterHMR/src
sh run.sh

Results will be saved in CenterHMR/demo/images_results.

Internet images

You can also run the code on random internet images via putting the images under CenterHMR/demo/images before running sh run.sh.

Or please refer to config_guide.md for detail configurations.

Please refer to config_guide.md for saving the estimated mesh/Center maps.

TODO LIST

The code will be gradually open sourced according to:

• the schedule
  • demo code for internet images or videos
  • evaluation code for re-implementation the results on 3DPW Challenge (really close)
  • runtime optimization

Citation

Please considering citing

@inproceedings{CenterHMR,
  title = {CenterHMR: a Bottom-up Single-shot Method for Multi-person 3D Mesh Recovery from a Single Image},
  author = {Yu, Sun and Qian, Bao and Wu, Liu and Yili, Fu and Tao, Mei},
  booktitle = {arxiv:2008.12272},
  month = {August},
  year = {2020}
}

Acknowledgement

We thank Peng Cheng for his constructive comments on Center map training.

Here are some great resources we benefit:

• SMPL models and layer is from SMPL-X model.
• Some functions are borrowed from HMR-pytorch.
• Some functions for data augmentation are borrowed from SPIN.
• Synthetic occlusion is borrowed from synthetic-occlusion
• The evaluation code of 3DPW dataset is brought from 3dpw-eval.
• For fair comparison, the GT annotations of 3DPW dataset are brought fromVIBE