This repository contains the dataset used in the work "Automatic Dance Video Segmentation for Understanding Choreography," which is accepted in MOCO '24.
A dataset was constructed by manually annotating segmentation points for 1410 videos in the AIST Dance DB.
The annotation data for basic videos are contained in basic_*.csv, and for advanced videos in advanced_*.csv.
The file *_count.csv represents segmentation points in counts (= musical beats), and the file *_sec.csv in seconds. You can use either one.
Note
The count starts with 1. That is, the first beat in the video corresponds to count 1.
For each file, the first row is the header, and each remaining row corresponds to each annotation data. For each row, the first column is the video URL, the second column is the annotator ID, and the remaining columns represent the annotated segmentation points.
All the contents in this repository can only be used for academic research purposes. Please cite our paper when you use it.