Stage1: Pretraining

Install environment:

python setup.py develop

Please download the SMPL models (version 1.0.0 compatible with Python 2.7, featuring 10 shape PCs) and relocate them to the corresponding locations:

mkdir lib/smpl/smpl_model/
mv /path/to/smpl/models/basicModel_f_lbs_10_207_0_v1.0.0.pkl lib/smpl/smpl_model/SMPL_NEUTRAL.pkl

Prepare Training Data

We combine THuman2.0 (526 scans) and CAPE (3, 000 scans) to train our joint two-layer implicit model.

1. THuman 2.0 dataset

i) Download THuman2.0 dataset following their instructions.

ii) Also download the corresponding SMPL parameters:

wget https://dataset.ait.ethz.ch/downloads/gdna/THuman2.0_smpl.zip
unzip THuman2.0_smpl.zip -d data/

iii) Run the pre-processing script to get ground truth occupancy:

python preprocess_thuman_data.py

2. CAPE dataset

i) Download CAPES dataset following their instructions.

ii) Run the pre-processing script to get ground truth occupancy:

python preprocess_cape_data.py

python train.py expname=pretrain