Tracking and tracing audiovisual reuse: Introducing the Video Reuse Detector

The reuse and reappropriation of audiovisual content have been a recurring topic of research in the humanities, not least in studies of remix cultures. An open question that remains, however, is how artificial intelligence and machine learning may help scholars study the reuse of audiovisual heritage. In this article, we introduce the Video Reuse Detector (VRD) – a methodological toolkit for identifying visual similarities in audiovisual archives with the help of machine learning. Designed to assist in the study of the “social life” and “cultural biographies” ( Kopytoff 1986 , Appadurai 1986 ) of video clips, the VRD helps explore how the meaning of historic footage changes when it circulates and is recycled/cross-referenced in video productions through time. The toolkit uses machine learning techniques (specifically, convolutional neural networks) combined with tools for performing similarity searches (specifically, the Faiss library) to detect copies in audiovisual archives. It also assembles a series of tools for trimming and preparing datasets and filtering/visualizing matching results. Inspired by the “visual turn” in digital history and digital humanities research, the article introduces and exemplifies the basic logic and rationale behind the VRD, and discusses how the digitization of audiovisual archives opens new ways of exploring the reuse of historic moving images.