BRACE

The Breakdancing Competition Dataset for Dance Motion Synthesis

Dataset Information
Modalities
Videos, Audio, Point cloud, Actions
Languages
English
Introduced
2022
License
S-Lab License 1.0
Homepage
Official Website
Contents

Overview

BRACE is a dataset for audio-conditioned dance motion synthesis challenging common assumptions for this task:

  • strong music-dance correlation
  • controlled motion data
  • simple poses and movements

To address these issues:

  • We focus on breakdancing which features acrobatic moves, tangled postures and weaker dance-music correlation.
  • We adopt a hybrid labelling pipeline leveraging estimation models as well as manual annotations to obtain good quality keypoint sequences at a reduced cost.
  • Our efforts produced the BRACE dataset, which contains over 3 hours and 30 minutes of densely annotated poses.
  • BRACE is also useful to fine-tune pose-estimation models thanks to its high quality keypoint annotations for complicated and uncommon poses.

Variants: BRACE

Associated Benchmarks

This dataset is used in 2 benchmarks:

  • Pose Estimation - Metrics: Average Precision, Average Recall
  • Motion Synthesis - Metrics: Frechet Inception Distance, Beat alignment score, Beat DTW cost, Footwork average, Powermove average, Toprock average

Recent Benchmark Submissions

Task Model Paper Date
Motion Synthesis AIST++ AI Choreographer: Music Conditioned 3D … 2021-01-21
Motion Synthesis Dance Revolution Dance Revolution: Long-Term Dance Generation … 2020-06-11
Motion Synthesis Dancing 2 Music Dancing to Music 2019-11-05
Pose Estimation HRNet fine-tuned on BRACE Deep High-Resolution Representation Learning for … 2019-02-25
Pose Estimation HRNet pre-trained on COCO Deep High-Resolution Representation Learning for … 2019-02-25

Research Papers

Recent papers with results on this dataset:

External Links: