Design a site like this with
Get started

Dance Documentation and Computer Vision

Mesh overlay on historical dance documentation.
Image credit: “Dunham Technique: Dunham walk—slow.” Video. ; Modeling: Nicola Plant / Visceral Histories, Visual Arguments: Dance-Based Approaches to Data.

We are knee-deep into Visceral Histories, Visual Arguments, which has unexpectedly led us deeper into emerging practices in computer vision and machine learning. At the end of January, we’ll be discussing some of the questions arising from this new research at the Center for Dance Research (C-DaRE) in a conversation with Simon Ellis and Kevin Walker that we titled: “If the Archive Can’t Consent: Historical Dance Footage, Computer Vision, and the Ethics of AI” (update: recording below). The questions we are bringing are:

  • What happens when we engage historical dance footage as a data source? 
  • If the archive can’t consent, how might we propose a method of analysis that is not based on a politics of capture?
  • How do historical materials defamiliarize norms of extraction, and dominant ideologies of the body on which models tend to be trained, in particular with regard to minoritarian subjects? 
  • What would it take for AI-driven computer vision to be informed by specific dance-based ways of knowing embodiment? 

Alongside our work with the Institute for Dunham Technique Certification (IDTC) on data that comes from the dance studio, team-member Nicola Plant has been researching the potential for computer vision to uncover movement data from historical dance documentation. In the process, we have come to realize that many of the available tools for pose estimation have a rigid spinal structure, and are therefore not equipped to support the analysis of dance forms–such as African and Afro-diasporic dances–where spinal articulation is central to the body’s movement.

In support of this research strand, Harmony and Ohio State colleagues Vita Berezina-Blackburn, Crystal Michelle Perkins, Roger Crawfish, and graduate researcher Mirkamil Mierkamili were awarded a research grant to engage in a culturally-aware process working toward the development of a pose estimation model better attuned to Black dance practices. The OSU folks got to start playing with some motion capture, and Rachel Tavernier and Celia Benvenutti of IDTC will join Kate and Harmony at OSU for a brief residency to engage in this experimental work. We’re excited to see what emerges!

UPDATE: The full presentation and discussion for If the Archive Can’t Consent is now online. You can see the above Dunham Walk frame in action.

%d bloggers like this: