Getting started

This code has been assembled from the following 3 repositories:

COCO to SMPL Estimator: https://github.com/google/aistplusplus_api
Visualize 3D Keypoints: https://github.com/Daniil-Osokin/lightweight-human-pose-estimation-3d-demo.pytorch
SMPL-to-FBX: https://github.com/softcat477/SMPL-to-FBX.git

The code has been tested on Ubuntu 20.04 and requires:

Python 3.7
Conda
CUDA capable GPU

Setup for keypoints visualizer

Set up the Anaconda environment:

conda create -n VisDemo python=3.7
conda activate VisDemo
setup_visualizer.sh

Visualize 3D Keypoints

cd lightweight-human-pose-estimation-3d-demo.pytorch 
python plot_AIST_Keypoints.py --joints ../aist_sample.npy

COCO 3D Body Keypoints to SMPL Estimation

conda create -n COCO_Estimator python=3.7
conda activate COCO_Estimator
setup_coco.sh

Further steps:

Download SMPL body model: https://download.is.tue.mpg.de/download.php?domain=smpl&sfile=SMPL_python_v.1.0.0.zip and place the "m" body as SMPL_MALE.pkl in COCO_to_SMPL_Estimation/smpl_body/
Replace aist_plusplus_api-1.1.0-py3.7.egg in the created Anaconda env in site-packages/ with the one provided in this repository
Place the aist_sample.pkl in the COCO_to_SMPL_Estimation/keypoints_dir/keypoints3d

Command to estimate:

python processing/run_estimate_smpl.py --anno_dir ./keypoints_dir/ --smpl_dir ./smpl_body/ --save_dir ./keypoints_dir/motions/ --data_type internal

SMPL Motion to FBX

conda create -n FBX python=3.7
conda activate FBX
setup_fbx.sh

Setting up Python FBX:

Download Python FBX SDK: https://www.autodesk.com/developer-network/platform-technologies/fbx-sdk-2020-3
Extract from the folder
Run and extract the FBX files into your directory (Eg. MY_FBX_FILES)
Copy the files to site-packages of the conda env