Swing Dance Dataset

dancing_-A4SnMrlK88_0_200_grid1x2.mov

Swing Dance dataset is a large-scale dataset comprising about 30 hours of swing dance videos with 680 unique couple subjects represented as 3D humans. We lifted each video into 4D using SLAHMR to extract the 3D human motion in the world. This data was used for unary and dyadic human motion prediction. For more info please see our synNsync project website.

Approach 1: Manual Set up

1. Create a folder with the following structure:

   • swing_data/raw_data/ - contains unprocessed web scraped videos
   • swing_data/processed_data/ - contains processed videos with trimmed intro and outros
   • swing_data/slahmr - contains .npz files of the 4D humans annotations of the processed data

2. Download the youtube videos using your choice of method into swing_data/raw_data/ by using the provided list of video links found in video_links.txt. Each video can be found by replacing video_link_id in https://www.youtube.com/watch?v=video_link_id with the link found in the video_links.txt

   • Example:

     https://www.youtube.com/watch?v=laLyyC_RXa4
     

3. After the download, run the ffmpeg commands in the ffmpeg_commands.txt file to process the videos and save video into the swing_data/processed_data/ folder

4. Download SLAHMR output files from this URL and unzip into swing_data/slahmr

Approach 2: Automatic Set up

1. Install required packages

   pip install pytubefix
   pip install gdown
   

2. Run code to download and process the swing dance dataset

   python get_swing_data.py 
   

Swing Dance SLAHMR output (4D humans annotations):

• The naming convention of the files are : swing_data/slahmr/dancing_<video_link>_<start_frame>_<end_frame>.npz
• Size of the Data: 2.43 GB

License

See License.