DOWNLOAD.md

Download

1. Create folders that store pretrained models, datasets, and predictions.

   export REPO_DIR=$PWD
   mkdir -p $REPO_DIR/models       # pre-trained models
   mkdir -p $REPO_DIR/models/bstro # pre-trained models

2. Download pretrained models.

   First, download the pre-trained weights from the website and place it under models/bstro.

   Second, run the following script.

   cd $REPO_DIR
   bash scripts/download_models.sh

   The scripts will download the pretrained METRO models that are trained for mesh reconstruction on 3DPW. For your convenience, this script will also download HRNet pre-trained weights, which will be used in training.

   The resulting data structure should follow the hierarchy as below.

   ${REPO_DIR}  
   |-- models  
   |   |-- metro_release
   |   |   |-- metro_3dpw_state_dict.bin
   |   |-- hrnet
   |   |   |-- hrnetv2_w40_imagenet_pretrained.pth
   |   |   |-- hrnetv2_w64_imagenet_pretrained.pth
   |   |   |-- cls_hrnet_w40_sgd_lr5e-2_wd1e-4_bs32_x100.yaml
   |   |   |-- cls_hrnet_w64_sgd_lr5e-2_wd1e-4_bs32_x100.yaml
   |-- metro 
   |-- datasets 
   |-- predictions 
   |-- README.md 
   |-- ... 
   |-- ... 
   

3. Download SMPL models

   To run our code smoothly, please visit the following websites to download SMPL models.

   • Download basicModel_neutral_lbs_10_207_0_v1.0.0.pkl from SMPLify, and place it under ${REPO_DIR}/metro/modeling/data.

   Please put the downloaded files under the ${REPO_DIR}/metro/modeling/data directory. The data structure should follow the hierarchy below.

   ${REPO_DIR}  
   |-- metro  
   |   |-- modeling
   |   |   |-- data
   |   |   |   |-- basicModel_neutral_lbs_10_207_0_v1.0.0.pkl
   |-- models
   |-- datasets
   |-- predictions
   |-- README.md 
   |-- ... 
   |-- ... 
   

   Please check /metro/modeling/data/README.md for further details.