install.md

Installation

• [~5min] The basic installation, including pytorch, other dependency, etc. We use pytorch 1.8.1, cuda version 10.2, pytorch-lightning 1.2 (there will be some incompatible issue about slurm preemption for version above 1.2)

  git clone https://github.com/JudyYe/ihoi.git
  cd ihoi
  conda env create -f docs/env.yaml
  conda activate ihoi
  

• [~3min] Install other third-party libraries including: FrankMocap with my minor modification, MANO, Detectron2 and download our pretrained models.

  • Run the following script

  sh scripts/one_click.sh
  

  • If any step doesn't work, please follow the instruction from the official website.

• Setting SMPL-X/MANO Models

  • We use use SMPL-X and MANO model as 3D pose estimation output. You have to accept the license and download them from the original website.

  • Download SMPLX Model (Neutral model: SMPLX_NEUTRAL.pkl):

    • You can use SMPL-X model for body mocap instead of SMPL model.
    • Download SMPLX_NEUTRAL.pkl in the original SMPL website. You need to register to download the SMPLX data.
    • Put the SMPLX_NEUTRAL.pkl file in: ./externals/frankmocap/extra_data/smpl/SMPLX_NEUTRAL.pkl
    • This is for hand module and whole body module

  • Download MANO Model (Neutral model: MANO_LEFT.pkl, MANO_RIGHT.pkl):

    • Download Models & Code in the original MANO website. You need to register to download the MANO data.
    • Put the models/MANO_LEFT.pkl models/MANO_RIGHT.pkl file in: ./externals/mano/

  • Once you sucessfully installed and downloaded all, you should have the following files in your directory, here is a list of key files:

    ./externals
    |── detectron2
    |── mano
    |   |── MANO_LEFT.pkl    
    |   └── MANO_RIGHT.pkl
    └── frankmocap
        └── extra_data
            └── smpl
                |── basicModel_neutral_lbs_10_207_0_v1.0.0.pkl
                └── SMPLX_NEUTRAL.pkl
    
    

Acknowledgement

Structure of this installation instruction is inspired from FrankMocap.