Ballroom Dance Dataset

About data

Ballroom dance dataset contains wearable sensor data, video, and body keypoints from video of ballroom dance performance involving 7 dancers, 13 ballroom dance step types over 100 minutes. Every participant weared six inertial sensors on the elbows, hips, and knees, and performed a dance performance. We prepared a high speed video camera and shot videos of the performances. Every participant performed 20 times and we shot 10 times from the front of dancers, then the other 10 from the back. Also, the order of dance steps in the performance is same among the participants.
There are 13 types of ballroom dance steps in our performance data. We named each step type as Step A (Open Basic), Step B (Foot replace), Step C (Fan), Step D (Hocky Stick), Step E (Newyork to Right), Step F (Newyork to Left), Step G (Spot Turn), Step H (Natural Top), Step I (Opening Out), Step J (Alemana), Step K (Hand to Hand to Right), Step L (Hand to Hand to Left), Step M (Aida) for easy understanding. The characteristics of each step type can be seen in Ballroom Guide. Be careful that the definition of dance steps and names are sometimes different among dance councils. In this work, I refered Ballroom Guide, but you can choose any other guidelines and step names.


You can download the files from Sample data (180 MB) which contains less data (4 step types, 1 dancer) is also available.
You can use the dataset for research purposes only.

Data size

The size of each data is as follows.

Folder Size
wearable_sensor 113 MB
video 100 GB
keypoint 685 MB

Data Contents

You can download,, and Each of them includes data of 7 dancers. Wearable sensor data consists of 20 csv files which contains acc, gyr and timestamp data. Video data includes 20 mp4 data for each dancer and each of them is about 50 second length. Before the performance, you can see the dancer jumps. It can help to synchronize video and wearable sensor data if you need. Keypoints data includes the body parts location from the video data as json file using OpenPose.

Wearable sensor

[Size: 293 MB, Number of csv files: 7 dancers x 20 times x 6 wearable sensors = 840]
CSV file that contains the time variation data of accelerometer and gyroscope sampled in every 8ms.
There are six CSV files for each one dance performance: Left ankle, right ankle, left hip, right hip, left arm, and right arm like shown in the photos.


[Size: 101 GB, Number of videos: 7 dancers x 20 times = 140]
We shot 20 times for each dancer's performance by the high speed camera(FPS: 120).
First 10 from the front and the other 10 from the back as shown in the pictures above. The picture below shows the contents in "video/dancer1/" as an example.


[Size: 4 GB, Number of folders: 7 dancers x 20 times = 140 (Each folder includes json files as many as the corresponding video frame)]
From each performance video, body parts (keypoints) location data per a frame in video are obtained using Open Pose like the image.


There are seven dancers who participated in the data collection. Their height and extent of dance experience are summarized in the table below.
participant height(cm) experience(year)
dancer1 173 17
dancer2 176 5
dancer3 175 3
dancer4 171 4
dancer5 160 1
dancer6 182 1
dancer7 177 5