Scores for the Virtual
A research project on the creation of AI-based dance notations
about the piece
The project deals with the question of whether and how machine-based learning methods can be used to create an algorithm that is capable of developing dance scores.
It is planned for several years and will be continued by the company after the research phase.
We started by looking at which of the publicly available data sets would be suitable for this. For example, there is the UCF 101 set from the University of Central Florida. This data is based on the analysis of videos, just like the Kinetics 600 collection from Google. For the latter collection, over 500,000 dance videos were analysed and divided into different categories.
Since most of the data sets are very different and can hardly be combined in a meaningful way, we first decided on a specific choreographic language and chose the German-English choreographer Ernest Berk.
The company reconstructed his work a few years ago. There were no film recordings, only photos and some movement sequences by former dancers. The dancers then recreated the corresponding movement phrases from these materials.
We now want to process this material for the algorithm so that it ultimately faces the same task as the company did back then.
First, we created a photo set of about 1000 photos. We then processed these as a test with the StyleGan algorithm.
The next step will be to record the dancers’ material using motion capture and then combine the two sets.
Supported by the NATIONAL PERFORMANCE NETWORK - STEPPING OUT, funded by the Federal Government Commissioner for Culture and the Media as part of the NEUSTART KULTUR initiative. Aid Programme Dance.