/Ego-STAN

Pytorch Implementation of xREgoPose

Primary LanguagePython

Setup

Pytorch Lightning Implementation of xREgoPose

How to setup:

  1. Setup Virtualenv on your home directory.
cd ~/
module load python/3.9
virtualenv --no-download torch
source torch/bin/activate
pip install --no-index --upgrade pip
  1. Install required dependencies for the repo
# First cd into the xREgoPose directory 
# For example, I will have to " cd ~/projects/def-pfieguth/j97park/xREgoPose/
pip install -r requirements.txt

Navigation

  • train.py
    • Main wrapper for training
    • Options for training can be found inside the script
    • Testing can be done after training is finished. Check the arguments.
  • eval.py
    • Given a trained checkpoint, run the evaluation only.
  • net/
    • Folder that contains all the different Pytorch Lightning classes of different methods.
  • visualizations/
    • Folder that contains the visualizations included in the main paper.
  • results/
    • Folder that contains the csv files of results.
  • Legacy/
    • Folder that contains the codes before migrating to pytorch lightning.
  • dataset/
    • Folder that contains the codes for processing data.
  • data/
    • Folder that contains information about mo2cap2 (in progress) and a config file for xREgoPose.
  • base/
    • Folder that contains base class for dataset, eval and transform.
  • utils/
    • Folder that contains codes for utility. (Logging, metrics, configuration of joints etc.)
  • sbash_scripts/
    • Example scripts to run on compute canada.