-
Create folders that store body models, datasets, and logs.
export REPO_DIR=$PWD mkdir -p $REPO_DIR/body_models # body models mkdir -p $REPO_DIR/datasets # datasets mkdir -p $REPO_DIR/logs # logs
-
Download body models.
Get yourself registered and download
models_smplx_v1_1.zip
from SMPL-X, and place it at${REPO_DIR}/body_models/smplx
. The data structure should follow the hierarchy below.${REPO_DIR} |-- body_models | |-- smplx | | |-- SMPLX_NEUTRAL.pkl | | |-- SMPLX_NEUTRAL.npz | | |-- ... |-- datasets |-- logs |-- README.md |-- ...
-
Download motion retrieval datasets.
- Register and download all the available SMPL+H sequences from AMASS (This would take long time).
- Register and download
babel_v1-0_release.zip
from BABEL. Note that BABEL provides the annotations for the AMASS dataset.
Please place the datasets at
${REPO_DIR}/datasets
.You also need to download our manually preprocessed BABEL raw labels,
cd ${REPO_DIR}/datasets wget https://www.dropbox.com/s/6p69k1tl1gk4lom/clip-actor_preprocess.zip unzip clip-actor_preprocess.zip
Finally, you need to download
featp_2_fps.json
from BABEL github.The data structure should follow the hierarchy below.
${REPO_DIR} |-- body_models |-- datasets | |-- amass | | |-- ACCAD | | |-- BMLhandball | | |-- ... | | |-- ... | |-- babel | | |-- train.json | | |-- val.json | | |-- ... | | |-- raw_label.npy | | |-- encoded_raw_label.npy | | |-- featp_2_fps.json |-- logs |-- README.md |-- ...