Harmony4D Dataset Preparation

Follow the steps below to prepare the Harmony4D Dataset for SAM-3D-Body Data.

---

• Set the following environment variables to simplify directory references:

  export HARMONY4D_DATA_DIR=/path/to/harmony4d/dataset
  export HARMONY4D_IMG_DIR=/path/to/harmony4d/undistorted/images
  

• Download 🔗 Harmony4D Dataset and extract all the files to $HARMONY4D_DATA_DIR, following the directory structure below.

  $HARMONY4D_DATA_DIR
  ├── test
  │   ├── 01_hugging
  │   │   └── 002_hugging
  │   ├── 03_grappling2
  │   │   ├── 025_grappling2
  │   │   └── ...
  │   ├── ...
  └── train
      ├── 01_hugging
      │   └── 001_hugging
      ├── 02_grappling
      │   ├── 001_grappling
      │   └── ...
      └── ...
  

• Run the following command to undistort the Harmony4D images and save the results to $HARMONY4D_IMG_DIR.

  python scripts/harmony4d/undistort_harmony4d.py \
      --src_dir $HARMONY4D_DATA_DIR \
      --dst_dir $HARMONY4D_IMG_DIR
  

• $HARMONY4D_IMG_DIR should the directory structure below.

  $HARMONY4D_IMG_DIR
  ├── test
  │   ├── 01_hugging
  │   │   └── 002_hugging
  │   ├── 03_grappling2
  │   │   ├── 025_grappling2
  │   │   └── ...
  │   ├── ...
  └── train
      ├── 01_hugging
      │   └── 001_hugging
      ├── 02_grappling
      │   ├── 001_grappling
      │   └── ...
      └── ...