dataset_preparation.md

Data Preparation

In this document we describe the data involved in XRMoGen and how to use it.

• Data Preparation
  • Download Data
    • Data Structure

Download Data

The dataset used in the dance generation task is AIST++. It contains original music and motion annotations(https://google.github.io/aistplusplus_dataset/download.html). In XRMoGen, we provide extracted musical features (438 dimensions) and 3D coordinates of human key points extracted based on the SMPL model (download here).

Data Structure

We recommend using the extracted data. Please download it and unzip it to $PROJECT/data. In order to facilitate the synthesis of a video with music after the dance is generated, it is needed to download the original music (.mov) to the musics folder in the same directory:

xrmogen
├── mogen
├── docs
├── configs
├── data
│   ├── aistpp_train_wav
│   ├── aistpp_test_full_wav
│   ├── aistpp_music_feat_7.5fps
│   ├── aist_features_zero_start
│   ├── musics
├── ...