Clarinetist for Science (C4S) dataset

Fully-annotated dataset of audio-visual recordings of clarinetists performing classical pieces fragments. The annotations include manually cleaned note onsets (+36k notes), ~4.5h of recordings and visual annotations (face bounding box, face landmarks, wrists and clarinet coordinates). This dataset can be used for multimodal and cross-modal note onsets detection, body pose estimation, clarinet segmentation, and ancillary/expressive movements analysis.

 
example of ROIs sequence
 

The authors encourage research by publicly releasing recordings and annotations. The material is shared under the Attribution-NonCommercial-ShareAlike 4.0 International license (see https://creativecommons.org/licenses/by-nc-sa/4.0/).

 
THE DOWNLOAD LINKS WILL BE AVAILABLE FROM THE 4th of June 2017
 

Using the dataset? Please cite the following paper:

@conference {Bazzica2017,
  title = {Vision-based Detection of Acoustic Timed Events:
  a Case Study on Clarinet Note Onsets},
  booktitle = {International Workshop on Deep Learning for Music (DLM)
  in conjunction with the International Joint Conference on Neural Networks (IJCNN)},
  year = {2017},
  address = {Anchorage, Alaska (USA)},
  author = {Alessio Bazzica and Jan C van Gemert and Cynthia C.S. Liem and Alan Hanjalic}
}