directory.md

Directory

Root

The ${ROOT} is described as below.

${ROOT} 
|-- assset 
|-- data  
|-- experiment
|-- lib
|-- main  

• data contains required data and soft links to images and annotations directories.
• experiment contains log, trained models, visualized outputs.
• lib contains kernel codes for CycleAdapt.
• main contains high-level codes for training or testing the network.

Required data

You need to follow directory structure of the data as below.

${ROOT} 
|-- data  
|   |-- base_data
|   |   |-- human_models
|   |   |   |-- smpl
|   |   |   |   |-- SMPL_FEMALE.pkl
|   |   |   |   |-- SMPL_MALE.pkl
|   |   |   |   |-- SMPL_NEUTRAL.pkl
|   |   |   |-- J_regressor_h36m_smpl.npy
|   |   |   |-- smpl_mean_params.npz
|   |   |-- pose_prior
|   |   |   |-- gmm_08.pkl
|   |   |-- pretrained_models
|   |   |   |-- hmr_basemodel.pt
|   |   |   |-- md_basemodel.pt
|   |-- ...

• base_data/human_models contains smpl 3D model files. Download the SMPL model files from [smpl].
• base_data/pretrained_models contains HMRNet&MDNet checkpoints to be adapted.
• All files except smpl folder can be downloaded from [base_data].

Demo

To run CycleAdapt on a custom video, prepare parsed images and the annotation file as below.

${ROOT} 
|-- data  
|   |-- ...
|   |-- Demo
|   |   |-- images
|   |   |-- annotation.json
|   |-- ...

• To obtain parsed images from video, you can utilize python lib/utils/video2image.py --video {video_path}.
• To construct annotation file, we recommend running AlphaPose [codes] from the parsed images.
• We provide simple example, which can be downloaded from [here]

Dataset

${ROOT} 
|-- data  
|   |-- ...
|   |-- PW3D
|   |   |-- imageFiles
|   |   |-- 3DPW_test.json
|   |-- ...

• Download 3DPW parsed data [data] [annot]

Experiment

${ROOT}  
|-- experiment
|-- |-- {exp_name}  
|   |   |-- checkpoints  
|   |   |-- log  
|   |   |-- vis  

• Creating experiment folder as soft link form is recommended instead of folder form because it would take large storage capacity.
• checkpoints folder contains saved checkpoints after adaptation.
• log folder contains training log file.
• vis folder contains visualized results.